xref: /linux/arch/arm64/kvm/sys_regs.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  *
6  * Derived from arch/arm/kvm/coproc.c:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Authors: Rusty Russell <rusty@rustcorp.com.au>
9  *          Christoffer Dall <c.dall@virtualopensystems.com>
10  */
11 
12 #include <linux/bitfield.h>
13 #include <linux/bsearch.h>
14 #include <linux/cacheinfo.h>
15 #include <linux/debugfs.h>
16 #include <linux/kvm_host.h>
17 #include <linux/mm.h>
18 #include <linux/printk.h>
19 #include <linux/uaccess.h>
20 
21 #include <asm/arm_pmuv3.h>
22 #include <asm/cacheflush.h>
23 #include <asm/cputype.h>
24 #include <asm/debug-monitors.h>
25 #include <asm/esr.h>
26 #include <asm/kvm_arm.h>
27 #include <asm/kvm_emulate.h>
28 #include <asm/kvm_hyp.h>
29 #include <asm/kvm_mmu.h>
30 #include <asm/kvm_nested.h>
31 #include <asm/perf_event.h>
32 #include <asm/sysreg.h>
33 
34 #include <trace/events/kvm.h>
35 
36 #include "sys_regs.h"
37 #include "vgic/vgic.h"
38 
39 #include "trace.h"
40 
41 /*
42  * For AArch32, we only take care of what is being trapped. Anything
43  * that has to do with init and userspace access has to go via the
44  * 64bit interface.
45  */
46 
47 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
48 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
49 		      u64 val);
50 
51 static bool undef_access(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
52 			 const struct sys_reg_desc *r)
53 {
54 	kvm_inject_undefined(vcpu);
55 	return false;
56 }
57 
58 static bool bad_trap(struct kvm_vcpu *vcpu,
59 		     struct sys_reg_params *params,
60 		     const struct sys_reg_desc *r,
61 		     const char *msg)
62 {
63 	WARN_ONCE(1, "Unexpected %s\n", msg);
64 	print_sys_reg_instr(params);
65 	return undef_access(vcpu, params, r);
66 }
67 
68 static bool read_from_write_only(struct kvm_vcpu *vcpu,
69 				 struct sys_reg_params *params,
70 				 const struct sys_reg_desc *r)
71 {
72 	return bad_trap(vcpu, params, r,
73 			"sys_reg read to write-only register");
74 }
75 
76 static bool write_to_read_only(struct kvm_vcpu *vcpu,
77 			       struct sys_reg_params *params,
78 			       const struct sys_reg_desc *r)
79 {
80 	return bad_trap(vcpu, params, r,
81 			"sys_reg write to read-only register");
82 }
83 
84 #define PURE_EL2_SYSREG(el2)						\
85 	case el2: {							\
86 		*el1r = el2;						\
87 		return true;						\
88 	}
89 
90 #define MAPPED_EL2_SYSREG(el2, el1, fn)					\
91 	case el2: {							\
92 		*xlate = fn;						\
93 		*el1r = el1;						\
94 		return true;						\
95 	}
96 
97 static bool get_el2_to_el1_mapping(unsigned int reg,
98 				   unsigned int *el1r, u64 (**xlate)(u64))
99 {
100 	switch (reg) {
101 		PURE_EL2_SYSREG(  VPIDR_EL2	);
102 		PURE_EL2_SYSREG(  VMPIDR_EL2	);
103 		PURE_EL2_SYSREG(  ACTLR_EL2	);
104 		PURE_EL2_SYSREG(  HCR_EL2	);
105 		PURE_EL2_SYSREG(  MDCR_EL2	);
106 		PURE_EL2_SYSREG(  HSTR_EL2	);
107 		PURE_EL2_SYSREG(  HACR_EL2	);
108 		PURE_EL2_SYSREG(  VTTBR_EL2	);
109 		PURE_EL2_SYSREG(  VTCR_EL2	);
110 		PURE_EL2_SYSREG(  RVBAR_EL2	);
111 		PURE_EL2_SYSREG(  TPIDR_EL2	);
112 		PURE_EL2_SYSREG(  HPFAR_EL2	);
113 		PURE_EL2_SYSREG(  HCRX_EL2	);
114 		PURE_EL2_SYSREG(  HFGRTR_EL2	);
115 		PURE_EL2_SYSREG(  HFGWTR_EL2	);
116 		PURE_EL2_SYSREG(  HFGITR_EL2	);
117 		PURE_EL2_SYSREG(  HDFGRTR_EL2	);
118 		PURE_EL2_SYSREG(  HDFGWTR_EL2	);
119 		PURE_EL2_SYSREG(  HAFGRTR_EL2	);
120 		PURE_EL2_SYSREG(  CNTVOFF_EL2	);
121 		PURE_EL2_SYSREG(  CNTHCTL_EL2	);
122 		MAPPED_EL2_SYSREG(SCTLR_EL2,   SCTLR_EL1,
123 				  translate_sctlr_el2_to_sctlr_el1	     );
124 		MAPPED_EL2_SYSREG(CPTR_EL2,    CPACR_EL1,
125 				  translate_cptr_el2_to_cpacr_el1	     );
126 		MAPPED_EL2_SYSREG(TTBR0_EL2,   TTBR0_EL1,
127 				  translate_ttbr0_el2_to_ttbr0_el1	     );
128 		MAPPED_EL2_SYSREG(TTBR1_EL2,   TTBR1_EL1,   NULL	     );
129 		MAPPED_EL2_SYSREG(TCR_EL2,     TCR_EL1,
130 				  translate_tcr_el2_to_tcr_el1		     );
131 		MAPPED_EL2_SYSREG(VBAR_EL2,    VBAR_EL1,    NULL	     );
132 		MAPPED_EL2_SYSREG(AFSR0_EL2,   AFSR0_EL1,   NULL	     );
133 		MAPPED_EL2_SYSREG(AFSR1_EL2,   AFSR1_EL1,   NULL	     );
134 		MAPPED_EL2_SYSREG(ESR_EL2,     ESR_EL1,     NULL	     );
135 		MAPPED_EL2_SYSREG(FAR_EL2,     FAR_EL1,     NULL	     );
136 		MAPPED_EL2_SYSREG(MAIR_EL2,    MAIR_EL1,    NULL	     );
137 		MAPPED_EL2_SYSREG(TCR2_EL2,    TCR2_EL1,    NULL	     );
138 		MAPPED_EL2_SYSREG(PIR_EL2,     PIR_EL1,     NULL	     );
139 		MAPPED_EL2_SYSREG(PIRE0_EL2,   PIRE0_EL1,   NULL	     );
140 		MAPPED_EL2_SYSREG(POR_EL2,     POR_EL1,     NULL	     );
141 		MAPPED_EL2_SYSREG(AMAIR_EL2,   AMAIR_EL1,   NULL	     );
142 		MAPPED_EL2_SYSREG(ELR_EL2,     ELR_EL1,	    NULL	     );
143 		MAPPED_EL2_SYSREG(SPSR_EL2,    SPSR_EL1,    NULL	     );
144 		MAPPED_EL2_SYSREG(ZCR_EL2,     ZCR_EL1,     NULL	     );
145 		MAPPED_EL2_SYSREG(CONTEXTIDR_EL2, CONTEXTIDR_EL1, NULL	     );
146 	default:
147 		return false;
148 	}
149 }
150 
151 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
152 {
153 	u64 val = 0x8badf00d8badf00d;
154 	u64 (*xlate)(u64) = NULL;
155 	unsigned int el1r;
156 
157 	if (!vcpu_get_flag(vcpu, SYSREGS_ON_CPU))
158 		goto memory_read;
159 
160 	if (unlikely(get_el2_to_el1_mapping(reg, &el1r, &xlate))) {
161 		if (!is_hyp_ctxt(vcpu))
162 			goto memory_read;
163 
164 		/*
165 		 * CNTHCTL_EL2 requires some special treatment to
166 		 * account for the bits that can be set via CNTKCTL_EL1.
167 		 */
168 		switch (reg) {
169 		case CNTHCTL_EL2:
170 			if (vcpu_el2_e2h_is_set(vcpu)) {
171 				val = read_sysreg_el1(SYS_CNTKCTL);
172 				val &= CNTKCTL_VALID_BITS;
173 				val |= __vcpu_sys_reg(vcpu, reg) & ~CNTKCTL_VALID_BITS;
174 				return val;
175 			}
176 			break;
177 		}
178 
179 		/*
180 		 * If this register does not have an EL1 counterpart,
181 		 * then read the stored EL2 version.
182 		 */
183 		if (reg == el1r)
184 			goto memory_read;
185 
186 		/*
187 		 * If we have a non-VHE guest and that the sysreg
188 		 * requires translation to be used at EL1, use the
189 		 * in-memory copy instead.
190 		 */
191 		if (!vcpu_el2_e2h_is_set(vcpu) && xlate)
192 			goto memory_read;
193 
194 		/* Get the current version of the EL1 counterpart. */
195 		WARN_ON(!__vcpu_read_sys_reg_from_cpu(el1r, &val));
196 		if (reg >= __SANITISED_REG_START__)
197 			val = kvm_vcpu_apply_reg_masks(vcpu, reg, val);
198 
199 		return val;
200 	}
201 
202 	/* EL1 register can't be on the CPU if the guest is in vEL2. */
203 	if (unlikely(is_hyp_ctxt(vcpu)))
204 		goto memory_read;
205 
206 	if (__vcpu_read_sys_reg_from_cpu(reg, &val))
207 		return val;
208 
209 memory_read:
210 	return __vcpu_sys_reg(vcpu, reg);
211 }
212 
213 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
214 {
215 	u64 (*xlate)(u64) = NULL;
216 	unsigned int el1r;
217 
218 	if (!vcpu_get_flag(vcpu, SYSREGS_ON_CPU))
219 		goto memory_write;
220 
221 	if (unlikely(get_el2_to_el1_mapping(reg, &el1r, &xlate))) {
222 		if (!is_hyp_ctxt(vcpu))
223 			goto memory_write;
224 
225 		/*
226 		 * Always store a copy of the write to memory to avoid having
227 		 * to reverse-translate virtual EL2 system registers for a
228 		 * non-VHE guest hypervisor.
229 		 */
230 		__vcpu_sys_reg(vcpu, reg) = val;
231 
232 		switch (reg) {
233 		case CNTHCTL_EL2:
234 			/*
235 			 * If E2H=0, CNHTCTL_EL2 is a pure shadow register.
236 			 * Otherwise, some of the bits are backed by
237 			 * CNTKCTL_EL1, while the rest is kept in memory.
238 			 * Yes, this is fun stuff.
239 			 */
240 			if (vcpu_el2_e2h_is_set(vcpu))
241 				write_sysreg_el1(val, SYS_CNTKCTL);
242 			return;
243 		}
244 
245 		/* No EL1 counterpart? We're done here.? */
246 		if (reg == el1r)
247 			return;
248 
249 		if (!vcpu_el2_e2h_is_set(vcpu) && xlate)
250 			val = xlate(val);
251 
252 		/* Redirect this to the EL1 version of the register. */
253 		WARN_ON(!__vcpu_write_sys_reg_to_cpu(val, el1r));
254 		return;
255 	}
256 
257 	/* EL1 register can't be on the CPU if the guest is in vEL2. */
258 	if (unlikely(is_hyp_ctxt(vcpu)))
259 		goto memory_write;
260 
261 	if (__vcpu_write_sys_reg_to_cpu(val, reg))
262 		return;
263 
264 memory_write:
265 	 __vcpu_sys_reg(vcpu, reg) = val;
266 }
267 
268 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
269 #define CSSELR_MAX 14
270 
271 /*
272  * Returns the minimum line size for the selected cache, expressed as
273  * Log2(bytes).
274  */
275 static u8 get_min_cache_line_size(bool icache)
276 {
277 	u64 ctr = read_sanitised_ftr_reg(SYS_CTR_EL0);
278 	u8 field;
279 
280 	if (icache)
281 		field = SYS_FIELD_GET(CTR_EL0, IminLine, ctr);
282 	else
283 		field = SYS_FIELD_GET(CTR_EL0, DminLine, ctr);
284 
285 	/*
286 	 * Cache line size is represented as Log2(words) in CTR_EL0.
287 	 * Log2(bytes) can be derived with the following:
288 	 *
289 	 * Log2(words) + 2 = Log2(bytes / 4) + 2
290 	 * 		   = Log2(bytes) - 2 + 2
291 	 * 		   = Log2(bytes)
292 	 */
293 	return field + 2;
294 }
295 
296 /* Which cache CCSIDR represents depends on CSSELR value. */
297 static u32 get_ccsidr(struct kvm_vcpu *vcpu, u32 csselr)
298 {
299 	u8 line_size;
300 
301 	if (vcpu->arch.ccsidr)
302 		return vcpu->arch.ccsidr[csselr];
303 
304 	line_size = get_min_cache_line_size(csselr & CSSELR_EL1_InD);
305 
306 	/*
307 	 * Fabricate a CCSIDR value as the overriding value does not exist.
308 	 * The real CCSIDR value will not be used as it can vary by the
309 	 * physical CPU which the vcpu currently resides in.
310 	 *
311 	 * The line size is determined with get_min_cache_line_size(), which
312 	 * should be valid for all CPUs even if they have different cache
313 	 * configuration.
314 	 *
315 	 * The associativity bits are cleared, meaning the geometry of all data
316 	 * and unified caches (which are guaranteed to be PIPT and thus
317 	 * non-aliasing) are 1 set and 1 way.
318 	 * Guests should not be doing cache operations by set/way at all, and
319 	 * for this reason, we trap them and attempt to infer the intent, so
320 	 * that we can flush the entire guest's address space at the appropriate
321 	 * time. The exposed geometry minimizes the number of the traps.
322 	 * [If guests should attempt to infer aliasing properties from the
323 	 * geometry (which is not permitted by the architecture), they would
324 	 * only do so for virtually indexed caches.]
325 	 *
326 	 * We don't check if the cache level exists as it is allowed to return
327 	 * an UNKNOWN value if not.
328 	 */
329 	return SYS_FIELD_PREP(CCSIDR_EL1, LineSize, line_size - 4);
330 }
331 
332 static int set_ccsidr(struct kvm_vcpu *vcpu, u32 csselr, u32 val)
333 {
334 	u8 line_size = FIELD_GET(CCSIDR_EL1_LineSize, val) + 4;
335 	u32 *ccsidr = vcpu->arch.ccsidr;
336 	u32 i;
337 
338 	if ((val & CCSIDR_EL1_RES0) ||
339 	    line_size < get_min_cache_line_size(csselr & CSSELR_EL1_InD))
340 		return -EINVAL;
341 
342 	if (!ccsidr) {
343 		if (val == get_ccsidr(vcpu, csselr))
344 			return 0;
345 
346 		ccsidr = kmalloc_array(CSSELR_MAX, sizeof(u32), GFP_KERNEL_ACCOUNT);
347 		if (!ccsidr)
348 			return -ENOMEM;
349 
350 		for (i = 0; i < CSSELR_MAX; i++)
351 			ccsidr[i] = get_ccsidr(vcpu, i);
352 
353 		vcpu->arch.ccsidr = ccsidr;
354 	}
355 
356 	ccsidr[csselr] = val;
357 
358 	return 0;
359 }
360 
361 static bool access_rw(struct kvm_vcpu *vcpu,
362 		      struct sys_reg_params *p,
363 		      const struct sys_reg_desc *r)
364 {
365 	if (p->is_write)
366 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
367 	else
368 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
369 
370 	return true;
371 }
372 
373 /*
374  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
375  */
376 static bool access_dcsw(struct kvm_vcpu *vcpu,
377 			struct sys_reg_params *p,
378 			const struct sys_reg_desc *r)
379 {
380 	if (!p->is_write)
381 		return read_from_write_only(vcpu, p, r);
382 
383 	/*
384 	 * Only track S/W ops if we don't have FWB. It still indicates
385 	 * that the guest is a bit broken (S/W operations should only
386 	 * be done by firmware, knowing that there is only a single
387 	 * CPU left in the system, and certainly not from non-secure
388 	 * software).
389 	 */
390 	if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
391 		kvm_set_way_flush(vcpu);
392 
393 	return true;
394 }
395 
396 static bool access_dcgsw(struct kvm_vcpu *vcpu,
397 			 struct sys_reg_params *p,
398 			 const struct sys_reg_desc *r)
399 {
400 	if (!kvm_has_mte(vcpu->kvm))
401 		return undef_access(vcpu, p, r);
402 
403 	/* Treat MTE S/W ops as we treat the classic ones: with contempt */
404 	return access_dcsw(vcpu, p, r);
405 }
406 
407 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
408 {
409 	switch (r->aarch32_map) {
410 	case AA32_LO:
411 		*mask = GENMASK_ULL(31, 0);
412 		*shift = 0;
413 		break;
414 	case AA32_HI:
415 		*mask = GENMASK_ULL(63, 32);
416 		*shift = 32;
417 		break;
418 	default:
419 		*mask = GENMASK_ULL(63, 0);
420 		*shift = 0;
421 		break;
422 	}
423 }
424 
425 /*
426  * Generic accessor for VM registers. Only called as long as HCR_TVM
427  * is set. If the guest enables the MMU, we stop trapping the VM
428  * sys_regs and leave it in complete control of the caches.
429  */
430 static bool access_vm_reg(struct kvm_vcpu *vcpu,
431 			  struct sys_reg_params *p,
432 			  const struct sys_reg_desc *r)
433 {
434 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
435 	u64 val, mask, shift;
436 
437 	BUG_ON(!p->is_write);
438 
439 	get_access_mask(r, &mask, &shift);
440 
441 	if (~mask) {
442 		val = vcpu_read_sys_reg(vcpu, r->reg);
443 		val &= ~mask;
444 	} else {
445 		val = 0;
446 	}
447 
448 	val |= (p->regval & (mask >> shift)) << shift;
449 	vcpu_write_sys_reg(vcpu, val, r->reg);
450 
451 	kvm_toggle_cache(vcpu, was_enabled);
452 	return true;
453 }
454 
455 static bool access_actlr(struct kvm_vcpu *vcpu,
456 			 struct sys_reg_params *p,
457 			 const struct sys_reg_desc *r)
458 {
459 	u64 mask, shift;
460 
461 	if (p->is_write)
462 		return ignore_write(vcpu, p);
463 
464 	get_access_mask(r, &mask, &shift);
465 	p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
466 
467 	return true;
468 }
469 
470 /*
471  * Trap handler for the GICv3 SGI generation system register.
472  * Forward the request to the VGIC emulation.
473  * The cp15_64 code makes sure this automatically works
474  * for both AArch64 and AArch32 accesses.
475  */
476 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
477 			   struct sys_reg_params *p,
478 			   const struct sys_reg_desc *r)
479 {
480 	bool g1;
481 
482 	if (!kvm_has_gicv3(vcpu->kvm))
483 		return undef_access(vcpu, p, r);
484 
485 	if (!p->is_write)
486 		return read_from_write_only(vcpu, p, r);
487 
488 	/*
489 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
490 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
491 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
492 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
493 	 * group.
494 	 */
495 	if (p->Op0 == 0) {		/* AArch32 */
496 		switch (p->Op1) {
497 		default:		/* Keep GCC quiet */
498 		case 0:			/* ICC_SGI1R */
499 			g1 = true;
500 			break;
501 		case 1:			/* ICC_ASGI1R */
502 		case 2:			/* ICC_SGI0R */
503 			g1 = false;
504 			break;
505 		}
506 	} else {			/* AArch64 */
507 		switch (p->Op2) {
508 		default:		/* Keep GCC quiet */
509 		case 5:			/* ICC_SGI1R_EL1 */
510 			g1 = true;
511 			break;
512 		case 6:			/* ICC_ASGI1R_EL1 */
513 		case 7:			/* ICC_SGI0R_EL1 */
514 			g1 = false;
515 			break;
516 		}
517 	}
518 
519 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
520 
521 	return true;
522 }
523 
524 static bool access_gic_sre(struct kvm_vcpu *vcpu,
525 			   struct sys_reg_params *p,
526 			   const struct sys_reg_desc *r)
527 {
528 	if (!kvm_has_gicv3(vcpu->kvm))
529 		return undef_access(vcpu, p, r);
530 
531 	if (p->is_write)
532 		return ignore_write(vcpu, p);
533 
534 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
535 	return true;
536 }
537 
538 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
539 			struct sys_reg_params *p,
540 			const struct sys_reg_desc *r)
541 {
542 	if (p->is_write)
543 		return ignore_write(vcpu, p);
544 	else
545 		return read_zero(vcpu, p);
546 }
547 
548 /*
549  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
550  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
551  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
552  * treat it separately.
553  */
554 static bool trap_loregion(struct kvm_vcpu *vcpu,
555 			  struct sys_reg_params *p,
556 			  const struct sys_reg_desc *r)
557 {
558 	u32 sr = reg_to_encoding(r);
559 
560 	if (!kvm_has_feat(vcpu->kvm, ID_AA64MMFR1_EL1, LO, IMP))
561 		return undef_access(vcpu, p, r);
562 
563 	if (p->is_write && sr == SYS_LORID_EL1)
564 		return write_to_read_only(vcpu, p, r);
565 
566 	return trap_raz_wi(vcpu, p, r);
567 }
568 
569 static bool trap_oslar_el1(struct kvm_vcpu *vcpu,
570 			   struct sys_reg_params *p,
571 			   const struct sys_reg_desc *r)
572 {
573 	u64 oslsr;
574 
575 	if (!p->is_write)
576 		return read_from_write_only(vcpu, p, r);
577 
578 	/* Forward the OSLK bit to OSLSR */
579 	oslsr = __vcpu_sys_reg(vcpu, OSLSR_EL1) & ~OSLSR_EL1_OSLK;
580 	if (p->regval & OSLAR_EL1_OSLK)
581 		oslsr |= OSLSR_EL1_OSLK;
582 
583 	__vcpu_sys_reg(vcpu, OSLSR_EL1) = oslsr;
584 	return true;
585 }
586 
587 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
588 			   struct sys_reg_params *p,
589 			   const struct sys_reg_desc *r)
590 {
591 	if (p->is_write)
592 		return write_to_read_only(vcpu, p, r);
593 
594 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
595 	return true;
596 }
597 
598 static int set_oslsr_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
599 			 u64 val)
600 {
601 	/*
602 	 * The only modifiable bit is the OSLK bit. Refuse the write if
603 	 * userspace attempts to change any other bit in the register.
604 	 */
605 	if ((val ^ rd->val) & ~OSLSR_EL1_OSLK)
606 		return -EINVAL;
607 
608 	__vcpu_sys_reg(vcpu, rd->reg) = val;
609 	return 0;
610 }
611 
612 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
613 				   struct sys_reg_params *p,
614 				   const struct sys_reg_desc *r)
615 {
616 	if (p->is_write) {
617 		return ignore_write(vcpu, p);
618 	} else {
619 		p->regval = read_sysreg(dbgauthstatus_el1);
620 		return true;
621 	}
622 }
623 
624 /*
625  * We want to avoid world-switching all the DBG registers all the
626  * time:
627  *
628  * - If we've touched any debug register, it is likely that we're
629  *   going to touch more of them. It then makes sense to disable the
630  *   traps and start doing the save/restore dance
631  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
632  *   then mandatory to save/restore the registers, as the guest
633  *   depends on them.
634  *
635  * For this, we use a DIRTY bit, indicating the guest has modified the
636  * debug registers, used as follow:
637  *
638  * On guest entry:
639  * - If the dirty bit is set (because we're coming back from trapping),
640  *   disable the traps, save host registers, restore guest registers.
641  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
642  *   set the dirty bit, disable the traps, save host registers,
643  *   restore guest registers.
644  * - Otherwise, enable the traps
645  *
646  * On guest exit:
647  * - If the dirty bit is set, save guest registers, restore host
648  *   registers and clear the dirty bit. This ensure that the host can
649  *   now use the debug registers.
650  */
651 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
652 			    struct sys_reg_params *p,
653 			    const struct sys_reg_desc *r)
654 {
655 	access_rw(vcpu, p, r);
656 	if (p->is_write)
657 		vcpu_set_flag(vcpu, DEBUG_DIRTY);
658 
659 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
660 
661 	return true;
662 }
663 
664 /*
665  * reg_to_dbg/dbg_to_reg
666  *
667  * A 32 bit write to a debug register leave top bits alone
668  * A 32 bit read from a debug register only returns the bottom bits
669  *
670  * All writes will set the DEBUG_DIRTY flag to ensure the hyp code
671  * switches between host and guest values in future.
672  */
673 static void reg_to_dbg(struct kvm_vcpu *vcpu,
674 		       struct sys_reg_params *p,
675 		       const struct sys_reg_desc *rd,
676 		       u64 *dbg_reg)
677 {
678 	u64 mask, shift, val;
679 
680 	get_access_mask(rd, &mask, &shift);
681 
682 	val = *dbg_reg;
683 	val &= ~mask;
684 	val |= (p->regval & (mask >> shift)) << shift;
685 	*dbg_reg = val;
686 
687 	vcpu_set_flag(vcpu, DEBUG_DIRTY);
688 }
689 
690 static void dbg_to_reg(struct kvm_vcpu *vcpu,
691 		       struct sys_reg_params *p,
692 		       const struct sys_reg_desc *rd,
693 		       u64 *dbg_reg)
694 {
695 	u64 mask, shift;
696 
697 	get_access_mask(rd, &mask, &shift);
698 	p->regval = (*dbg_reg & mask) >> shift;
699 }
700 
701 static bool trap_bvr(struct kvm_vcpu *vcpu,
702 		     struct sys_reg_params *p,
703 		     const struct sys_reg_desc *rd)
704 {
705 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
706 
707 	if (p->is_write)
708 		reg_to_dbg(vcpu, p, rd, dbg_reg);
709 	else
710 		dbg_to_reg(vcpu, p, rd, dbg_reg);
711 
712 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
713 
714 	return true;
715 }
716 
717 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
718 		   u64 val)
719 {
720 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = val;
721 	return 0;
722 }
723 
724 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
725 		   u64 *val)
726 {
727 	*val = vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
728 	return 0;
729 }
730 
731 static u64 reset_bvr(struct kvm_vcpu *vcpu,
732 		      const struct sys_reg_desc *rd)
733 {
734 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = rd->val;
735 	return rd->val;
736 }
737 
738 static bool trap_bcr(struct kvm_vcpu *vcpu,
739 		     struct sys_reg_params *p,
740 		     const struct sys_reg_desc *rd)
741 {
742 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
743 
744 	if (p->is_write)
745 		reg_to_dbg(vcpu, p, rd, dbg_reg);
746 	else
747 		dbg_to_reg(vcpu, p, rd, dbg_reg);
748 
749 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
750 
751 	return true;
752 }
753 
754 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
755 		   u64 val)
756 {
757 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = val;
758 	return 0;
759 }
760 
761 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
762 		   u64 *val)
763 {
764 	*val = vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
765 	return 0;
766 }
767 
768 static u64 reset_bcr(struct kvm_vcpu *vcpu,
769 		      const struct sys_reg_desc *rd)
770 {
771 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = rd->val;
772 	return rd->val;
773 }
774 
775 static bool trap_wvr(struct kvm_vcpu *vcpu,
776 		     struct sys_reg_params *p,
777 		     const struct sys_reg_desc *rd)
778 {
779 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
780 
781 	if (p->is_write)
782 		reg_to_dbg(vcpu, p, rd, dbg_reg);
783 	else
784 		dbg_to_reg(vcpu, p, rd, dbg_reg);
785 
786 	trace_trap_reg(__func__, rd->CRm, p->is_write,
787 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm]);
788 
789 	return true;
790 }
791 
792 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
793 		   u64 val)
794 {
795 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = val;
796 	return 0;
797 }
798 
799 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
800 		   u64 *val)
801 {
802 	*val = vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
803 	return 0;
804 }
805 
806 static u64 reset_wvr(struct kvm_vcpu *vcpu,
807 		      const struct sys_reg_desc *rd)
808 {
809 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = rd->val;
810 	return rd->val;
811 }
812 
813 static bool trap_wcr(struct kvm_vcpu *vcpu,
814 		     struct sys_reg_params *p,
815 		     const struct sys_reg_desc *rd)
816 {
817 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
818 
819 	if (p->is_write)
820 		reg_to_dbg(vcpu, p, rd, dbg_reg);
821 	else
822 		dbg_to_reg(vcpu, p, rd, dbg_reg);
823 
824 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
825 
826 	return true;
827 }
828 
829 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
830 		   u64 val)
831 {
832 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = val;
833 	return 0;
834 }
835 
836 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
837 		   u64 *val)
838 {
839 	*val = vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
840 	return 0;
841 }
842 
843 static u64 reset_wcr(struct kvm_vcpu *vcpu,
844 		      const struct sys_reg_desc *rd)
845 {
846 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = rd->val;
847 	return rd->val;
848 }
849 
850 static u64 reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
851 {
852 	u64 amair = read_sysreg(amair_el1);
853 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
854 	return amair;
855 }
856 
857 static u64 reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
858 {
859 	u64 actlr = read_sysreg(actlr_el1);
860 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
861 	return actlr;
862 }
863 
864 static u64 reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
865 {
866 	u64 mpidr;
867 
868 	/*
869 	 * Map the vcpu_id into the first three affinity level fields of
870 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
871 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
872 	 * of the GICv3 to be able to address each CPU directly when
873 	 * sending IPIs.
874 	 */
875 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
876 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
877 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
878 	mpidr |= (1ULL << 31);
879 	vcpu_write_sys_reg(vcpu, mpidr, MPIDR_EL1);
880 
881 	return mpidr;
882 }
883 
884 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
885 				   const struct sys_reg_desc *r)
886 {
887 	if (kvm_vcpu_has_pmu(vcpu))
888 		return 0;
889 
890 	return REG_HIDDEN;
891 }
892 
893 static u64 reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
894 {
895 	u64 mask = BIT(ARMV8_PMU_CYCLE_IDX);
896 	u8 n = vcpu->kvm->arch.pmcr_n;
897 
898 	if (n)
899 		mask |= GENMASK(n - 1, 0);
900 
901 	reset_unknown(vcpu, r);
902 	__vcpu_sys_reg(vcpu, r->reg) &= mask;
903 
904 	return __vcpu_sys_reg(vcpu, r->reg);
905 }
906 
907 static u64 reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
908 {
909 	reset_unknown(vcpu, r);
910 	__vcpu_sys_reg(vcpu, r->reg) &= GENMASK(31, 0);
911 
912 	return __vcpu_sys_reg(vcpu, r->reg);
913 }
914 
915 static u64 reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
916 {
917 	/* This thing will UNDEF, who cares about the reset value? */
918 	if (!kvm_vcpu_has_pmu(vcpu))
919 		return 0;
920 
921 	reset_unknown(vcpu, r);
922 	__vcpu_sys_reg(vcpu, r->reg) &= kvm_pmu_evtyper_mask(vcpu->kvm);
923 
924 	return __vcpu_sys_reg(vcpu, r->reg);
925 }
926 
927 static u64 reset_pmselr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
928 {
929 	reset_unknown(vcpu, r);
930 	__vcpu_sys_reg(vcpu, r->reg) &= PMSELR_EL0_SEL_MASK;
931 
932 	return __vcpu_sys_reg(vcpu, r->reg);
933 }
934 
935 static u64 reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
936 {
937 	u64 pmcr = 0;
938 
939 	if (!kvm_supports_32bit_el0())
940 		pmcr |= ARMV8_PMU_PMCR_LC;
941 
942 	/*
943 	 * The value of PMCR.N field is included when the
944 	 * vCPU register is read via kvm_vcpu_read_pmcr().
945 	 */
946 	__vcpu_sys_reg(vcpu, r->reg) = pmcr;
947 
948 	return __vcpu_sys_reg(vcpu, r->reg);
949 }
950 
951 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
952 {
953 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
954 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
955 
956 	if (!enabled)
957 		kvm_inject_undefined(vcpu);
958 
959 	return !enabled;
960 }
961 
962 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
963 {
964 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
965 }
966 
967 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
968 {
969 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
970 }
971 
972 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
973 {
974 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
975 }
976 
977 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
978 {
979 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
980 }
981 
982 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
983 			const struct sys_reg_desc *r)
984 {
985 	u64 val;
986 
987 	if (pmu_access_el0_disabled(vcpu))
988 		return false;
989 
990 	if (p->is_write) {
991 		/*
992 		 * Only update writeable bits of PMCR (continuing into
993 		 * kvm_pmu_handle_pmcr() as well)
994 		 */
995 		val = kvm_vcpu_read_pmcr(vcpu);
996 		val &= ~ARMV8_PMU_PMCR_MASK;
997 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
998 		if (!kvm_supports_32bit_el0())
999 			val |= ARMV8_PMU_PMCR_LC;
1000 		kvm_pmu_handle_pmcr(vcpu, val);
1001 	} else {
1002 		/* PMCR.P & PMCR.C are RAZ */
1003 		val = kvm_vcpu_read_pmcr(vcpu)
1004 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
1005 		p->regval = val;
1006 	}
1007 
1008 	return true;
1009 }
1010 
1011 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1012 			  const struct sys_reg_desc *r)
1013 {
1014 	if (pmu_access_event_counter_el0_disabled(vcpu))
1015 		return false;
1016 
1017 	if (p->is_write)
1018 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
1019 	else
1020 		/* return PMSELR.SEL field */
1021 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
1022 			    & PMSELR_EL0_SEL_MASK;
1023 
1024 	return true;
1025 }
1026 
1027 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1028 			  const struct sys_reg_desc *r)
1029 {
1030 	u64 pmceid, mask, shift;
1031 
1032 	BUG_ON(p->is_write);
1033 
1034 	if (pmu_access_el0_disabled(vcpu))
1035 		return false;
1036 
1037 	get_access_mask(r, &mask, &shift);
1038 
1039 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
1040 	pmceid &= mask;
1041 	pmceid >>= shift;
1042 
1043 	p->regval = pmceid;
1044 
1045 	return true;
1046 }
1047 
1048 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
1049 {
1050 	u64 pmcr, val;
1051 
1052 	pmcr = kvm_vcpu_read_pmcr(vcpu);
1053 	val = FIELD_GET(ARMV8_PMU_PMCR_N, pmcr);
1054 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
1055 		kvm_inject_undefined(vcpu);
1056 		return false;
1057 	}
1058 
1059 	return true;
1060 }
1061 
1062 static int get_pmu_evcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1063 			  u64 *val)
1064 {
1065 	u64 idx;
1066 
1067 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 0)
1068 		/* PMCCNTR_EL0 */
1069 		idx = ARMV8_PMU_CYCLE_IDX;
1070 	else
1071 		/* PMEVCNTRn_EL0 */
1072 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1073 
1074 	*val = kvm_pmu_get_counter_value(vcpu, idx);
1075 	return 0;
1076 }
1077 
1078 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
1079 			      struct sys_reg_params *p,
1080 			      const struct sys_reg_desc *r)
1081 {
1082 	u64 idx = ~0UL;
1083 
1084 	if (r->CRn == 9 && r->CRm == 13) {
1085 		if (r->Op2 == 2) {
1086 			/* PMXEVCNTR_EL0 */
1087 			if (pmu_access_event_counter_el0_disabled(vcpu))
1088 				return false;
1089 
1090 			idx = SYS_FIELD_GET(PMSELR_EL0, SEL,
1091 					    __vcpu_sys_reg(vcpu, PMSELR_EL0));
1092 		} else if (r->Op2 == 0) {
1093 			/* PMCCNTR_EL0 */
1094 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
1095 				return false;
1096 
1097 			idx = ARMV8_PMU_CYCLE_IDX;
1098 		}
1099 	} else if (r->CRn == 0 && r->CRm == 9) {
1100 		/* PMCCNTR */
1101 		if (pmu_access_event_counter_el0_disabled(vcpu))
1102 			return false;
1103 
1104 		idx = ARMV8_PMU_CYCLE_IDX;
1105 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
1106 		/* PMEVCNTRn_EL0 */
1107 		if (pmu_access_event_counter_el0_disabled(vcpu))
1108 			return false;
1109 
1110 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1111 	}
1112 
1113 	/* Catch any decoding mistake */
1114 	WARN_ON(idx == ~0UL);
1115 
1116 	if (!pmu_counter_idx_valid(vcpu, idx))
1117 		return false;
1118 
1119 	if (p->is_write) {
1120 		if (pmu_access_el0_disabled(vcpu))
1121 			return false;
1122 
1123 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
1124 	} else {
1125 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
1126 	}
1127 
1128 	return true;
1129 }
1130 
1131 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1132 			       const struct sys_reg_desc *r)
1133 {
1134 	u64 idx, reg;
1135 
1136 	if (pmu_access_el0_disabled(vcpu))
1137 		return false;
1138 
1139 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
1140 		/* PMXEVTYPER_EL0 */
1141 		idx = SYS_FIELD_GET(PMSELR_EL0, SEL, __vcpu_sys_reg(vcpu, PMSELR_EL0));
1142 		reg = PMEVTYPER0_EL0 + idx;
1143 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
1144 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1145 		if (idx == ARMV8_PMU_CYCLE_IDX)
1146 			reg = PMCCFILTR_EL0;
1147 		else
1148 			/* PMEVTYPERn_EL0 */
1149 			reg = PMEVTYPER0_EL0 + idx;
1150 	} else {
1151 		BUG();
1152 	}
1153 
1154 	if (!pmu_counter_idx_valid(vcpu, idx))
1155 		return false;
1156 
1157 	if (p->is_write) {
1158 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
1159 		kvm_vcpu_pmu_restore_guest(vcpu);
1160 	} else {
1161 		p->regval = __vcpu_sys_reg(vcpu, reg);
1162 	}
1163 
1164 	return true;
1165 }
1166 
1167 static int set_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 val)
1168 {
1169 	bool set;
1170 
1171 	val &= kvm_pmu_accessible_counter_mask(vcpu);
1172 
1173 	switch (r->reg) {
1174 	case PMOVSSET_EL0:
1175 		/* CRm[1] being set indicates a SET register, and CLR otherwise */
1176 		set = r->CRm & 2;
1177 		break;
1178 	default:
1179 		/* Op2[0] being set indicates a SET register, and CLR otherwise */
1180 		set = r->Op2 & 1;
1181 		break;
1182 	}
1183 
1184 	if (set)
1185 		__vcpu_sys_reg(vcpu, r->reg) |= val;
1186 	else
1187 		__vcpu_sys_reg(vcpu, r->reg) &= ~val;
1188 
1189 	return 0;
1190 }
1191 
1192 static int get_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 *val)
1193 {
1194 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1195 
1196 	*val = __vcpu_sys_reg(vcpu, r->reg) & mask;
1197 	return 0;
1198 }
1199 
1200 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1201 			   const struct sys_reg_desc *r)
1202 {
1203 	u64 val, mask;
1204 
1205 	if (pmu_access_el0_disabled(vcpu))
1206 		return false;
1207 
1208 	mask = kvm_pmu_accessible_counter_mask(vcpu);
1209 	if (p->is_write) {
1210 		val = p->regval & mask;
1211 		if (r->Op2 & 0x1) {
1212 			/* accessing PMCNTENSET_EL0 */
1213 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
1214 			kvm_pmu_enable_counter_mask(vcpu, val);
1215 			kvm_vcpu_pmu_restore_guest(vcpu);
1216 		} else {
1217 			/* accessing PMCNTENCLR_EL0 */
1218 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
1219 			kvm_pmu_disable_counter_mask(vcpu, val);
1220 		}
1221 	} else {
1222 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
1223 	}
1224 
1225 	return true;
1226 }
1227 
1228 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1229 			   const struct sys_reg_desc *r)
1230 {
1231 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1232 
1233 	if (check_pmu_access_disabled(vcpu, 0))
1234 		return false;
1235 
1236 	if (p->is_write) {
1237 		u64 val = p->regval & mask;
1238 
1239 		if (r->Op2 & 0x1)
1240 			/* accessing PMINTENSET_EL1 */
1241 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
1242 		else
1243 			/* accessing PMINTENCLR_EL1 */
1244 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
1245 	} else {
1246 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1);
1247 	}
1248 
1249 	return true;
1250 }
1251 
1252 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1253 			 const struct sys_reg_desc *r)
1254 {
1255 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1256 
1257 	if (pmu_access_el0_disabled(vcpu))
1258 		return false;
1259 
1260 	if (p->is_write) {
1261 		if (r->CRm & 0x2)
1262 			/* accessing PMOVSSET_EL0 */
1263 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
1264 		else
1265 			/* accessing PMOVSCLR_EL0 */
1266 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
1267 	} else {
1268 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0);
1269 	}
1270 
1271 	return true;
1272 }
1273 
1274 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1275 			   const struct sys_reg_desc *r)
1276 {
1277 	u64 mask;
1278 
1279 	if (!p->is_write)
1280 		return read_from_write_only(vcpu, p, r);
1281 
1282 	if (pmu_write_swinc_el0_disabled(vcpu))
1283 		return false;
1284 
1285 	mask = kvm_pmu_accessible_counter_mask(vcpu);
1286 	kvm_pmu_software_increment(vcpu, p->regval & mask);
1287 	return true;
1288 }
1289 
1290 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1291 			     const struct sys_reg_desc *r)
1292 {
1293 	if (p->is_write) {
1294 		if (!vcpu_mode_priv(vcpu))
1295 			return undef_access(vcpu, p, r);
1296 
1297 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
1298 			       p->regval & ARMV8_PMU_USERENR_MASK;
1299 	} else {
1300 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
1301 			    & ARMV8_PMU_USERENR_MASK;
1302 	}
1303 
1304 	return true;
1305 }
1306 
1307 static int get_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1308 		    u64 *val)
1309 {
1310 	*val = kvm_vcpu_read_pmcr(vcpu);
1311 	return 0;
1312 }
1313 
1314 static int set_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1315 		    u64 val)
1316 {
1317 	u8 new_n = FIELD_GET(ARMV8_PMU_PMCR_N, val);
1318 	struct kvm *kvm = vcpu->kvm;
1319 
1320 	mutex_lock(&kvm->arch.config_lock);
1321 
1322 	/*
1323 	 * The vCPU can't have more counters than the PMU hardware
1324 	 * implements. Ignore this error to maintain compatibility
1325 	 * with the existing KVM behavior.
1326 	 */
1327 	if (!kvm_vm_has_ran_once(kvm) &&
1328 	    new_n <= kvm_arm_pmu_get_max_counters(kvm))
1329 		kvm->arch.pmcr_n = new_n;
1330 
1331 	mutex_unlock(&kvm->arch.config_lock);
1332 
1333 	/*
1334 	 * Ignore writes to RES0 bits, read only bits that are cleared on
1335 	 * vCPU reset, and writable bits that KVM doesn't support yet.
1336 	 * (i.e. only PMCR.N and bits [7:0] are mutable from userspace)
1337 	 * The LP bit is RES0 when FEAT_PMUv3p5 is not supported on the vCPU.
1338 	 * But, we leave the bit as it is here, as the vCPU's PMUver might
1339 	 * be changed later (NOTE: the bit will be cleared on first vCPU run
1340 	 * if necessary).
1341 	 */
1342 	val &= ARMV8_PMU_PMCR_MASK;
1343 
1344 	/* The LC bit is RES1 when AArch32 is not supported */
1345 	if (!kvm_supports_32bit_el0())
1346 		val |= ARMV8_PMU_PMCR_LC;
1347 
1348 	__vcpu_sys_reg(vcpu, r->reg) = val;
1349 	return 0;
1350 }
1351 
1352 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
1353 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
1354 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
1355 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
1356 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1357 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
1358 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1359 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
1360 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1361 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
1362 
1363 #define PMU_SYS_REG(name)						\
1364 	SYS_DESC(SYS_##name), .reset = reset_pmu_reg,			\
1365 	.visibility = pmu_visibility
1366 
1367 /* Macro to expand the PMEVCNTRn_EL0 register */
1368 #define PMU_PMEVCNTR_EL0(n)						\
1369 	{ PMU_SYS_REG(PMEVCNTRn_EL0(n)),				\
1370 	  .reset = reset_pmevcntr, .get_user = get_pmu_evcntr,		\
1371 	  .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
1372 
1373 /* Macro to expand the PMEVTYPERn_EL0 register */
1374 #define PMU_PMEVTYPER_EL0(n)						\
1375 	{ PMU_SYS_REG(PMEVTYPERn_EL0(n)),				\
1376 	  .reset = reset_pmevtyper,					\
1377 	  .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
1378 
1379 /* Macro to expand the AMU counter and type registers*/
1380 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
1381 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
1382 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
1383 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
1384 
1385 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1386 			const struct sys_reg_desc *rd)
1387 {
1388 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1389 }
1390 
1391 /*
1392  * If we land here on a PtrAuth access, that is because we didn't
1393  * fixup the access on exit by allowing the PtrAuth sysregs. The only
1394  * way this happens is when the guest does not have PtrAuth support
1395  * enabled.
1396  */
1397 #define __PTRAUTH_KEY(k)						\
1398 	{ SYS_DESC(SYS_## k), undef_access, reset_unknown, k,		\
1399 	.visibility = ptrauth_visibility}
1400 
1401 #define PTRAUTH_KEY(k)							\
1402 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1403 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1404 
1405 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1406 			      struct sys_reg_params *p,
1407 			      const struct sys_reg_desc *r)
1408 {
1409 	enum kvm_arch_timers tmr;
1410 	enum kvm_arch_timer_regs treg;
1411 	u64 reg = reg_to_encoding(r);
1412 
1413 	switch (reg) {
1414 	case SYS_CNTP_TVAL_EL0:
1415 	case SYS_AARCH32_CNTP_TVAL:
1416 		tmr = TIMER_PTIMER;
1417 		treg = TIMER_REG_TVAL;
1418 		break;
1419 	case SYS_CNTP_CTL_EL0:
1420 	case SYS_AARCH32_CNTP_CTL:
1421 		tmr = TIMER_PTIMER;
1422 		treg = TIMER_REG_CTL;
1423 		break;
1424 	case SYS_CNTP_CVAL_EL0:
1425 	case SYS_AARCH32_CNTP_CVAL:
1426 		tmr = TIMER_PTIMER;
1427 		treg = TIMER_REG_CVAL;
1428 		break;
1429 	case SYS_CNTPCT_EL0:
1430 	case SYS_CNTPCTSS_EL0:
1431 	case SYS_AARCH32_CNTPCT:
1432 		tmr = TIMER_PTIMER;
1433 		treg = TIMER_REG_CNT;
1434 		break;
1435 	default:
1436 		print_sys_reg_msg(p, "%s", "Unhandled trapped timer register");
1437 		return undef_access(vcpu, p, r);
1438 	}
1439 
1440 	if (p->is_write)
1441 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1442 	else
1443 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1444 
1445 	return true;
1446 }
1447 
1448 static s64 kvm_arm64_ftr_safe_value(u32 id, const struct arm64_ftr_bits *ftrp,
1449 				    s64 new, s64 cur)
1450 {
1451 	struct arm64_ftr_bits kvm_ftr = *ftrp;
1452 
1453 	/* Some features have different safe value type in KVM than host features */
1454 	switch (id) {
1455 	case SYS_ID_AA64DFR0_EL1:
1456 		switch (kvm_ftr.shift) {
1457 		case ID_AA64DFR0_EL1_PMUVer_SHIFT:
1458 			kvm_ftr.type = FTR_LOWER_SAFE;
1459 			break;
1460 		case ID_AA64DFR0_EL1_DebugVer_SHIFT:
1461 			kvm_ftr.type = FTR_LOWER_SAFE;
1462 			break;
1463 		}
1464 		break;
1465 	case SYS_ID_DFR0_EL1:
1466 		if (kvm_ftr.shift == ID_DFR0_EL1_PerfMon_SHIFT)
1467 			kvm_ftr.type = FTR_LOWER_SAFE;
1468 		break;
1469 	}
1470 
1471 	return arm64_ftr_safe_value(&kvm_ftr, new, cur);
1472 }
1473 
1474 /*
1475  * arm64_check_features() - Check if a feature register value constitutes
1476  * a subset of features indicated by the idreg's KVM sanitised limit.
1477  *
1478  * This function will check if each feature field of @val is the "safe" value
1479  * against idreg's KVM sanitised limit return from reset() callback.
1480  * If a field value in @val is the same as the one in limit, it is always
1481  * considered the safe value regardless For register fields that are not in
1482  * writable, only the value in limit is considered the safe value.
1483  *
1484  * Return: 0 if all the fields are safe. Otherwise, return negative errno.
1485  */
1486 static int arm64_check_features(struct kvm_vcpu *vcpu,
1487 				const struct sys_reg_desc *rd,
1488 				u64 val)
1489 {
1490 	const struct arm64_ftr_reg *ftr_reg;
1491 	const struct arm64_ftr_bits *ftrp = NULL;
1492 	u32 id = reg_to_encoding(rd);
1493 	u64 writable_mask = rd->val;
1494 	u64 limit = rd->reset(vcpu, rd);
1495 	u64 mask = 0;
1496 
1497 	/*
1498 	 * Hidden and unallocated ID registers may not have a corresponding
1499 	 * struct arm64_ftr_reg. Of course, if the register is RAZ we know the
1500 	 * only safe value is 0.
1501 	 */
1502 	if (sysreg_visible_as_raz(vcpu, rd))
1503 		return val ? -E2BIG : 0;
1504 
1505 	ftr_reg = get_arm64_ftr_reg(id);
1506 	if (!ftr_reg)
1507 		return -EINVAL;
1508 
1509 	ftrp = ftr_reg->ftr_bits;
1510 
1511 	for (; ftrp && ftrp->width; ftrp++) {
1512 		s64 f_val, f_lim, safe_val;
1513 		u64 ftr_mask;
1514 
1515 		ftr_mask = arm64_ftr_mask(ftrp);
1516 		if ((ftr_mask & writable_mask) != ftr_mask)
1517 			continue;
1518 
1519 		f_val = arm64_ftr_value(ftrp, val);
1520 		f_lim = arm64_ftr_value(ftrp, limit);
1521 		mask |= ftr_mask;
1522 
1523 		if (f_val == f_lim)
1524 			safe_val = f_val;
1525 		else
1526 			safe_val = kvm_arm64_ftr_safe_value(id, ftrp, f_val, f_lim);
1527 
1528 		if (safe_val != f_val)
1529 			return -E2BIG;
1530 	}
1531 
1532 	/* For fields that are not writable, values in limit are the safe values. */
1533 	if ((val & ~mask) != (limit & ~mask))
1534 		return -E2BIG;
1535 
1536 	return 0;
1537 }
1538 
1539 static u8 pmuver_to_perfmon(u8 pmuver)
1540 {
1541 	switch (pmuver) {
1542 	case ID_AA64DFR0_EL1_PMUVer_IMP:
1543 		return ID_DFR0_EL1_PerfMon_PMUv3;
1544 	case ID_AA64DFR0_EL1_PMUVer_IMP_DEF:
1545 		return ID_DFR0_EL1_PerfMon_IMPDEF;
1546 	default:
1547 		/* Anything ARMv8.1+ and NI have the same value. For now. */
1548 		return pmuver;
1549 	}
1550 }
1551 
1552 static u64 sanitise_id_aa64pfr0_el1(const struct kvm_vcpu *vcpu, u64 val);
1553 static u64 sanitise_id_aa64dfr0_el1(const struct kvm_vcpu *vcpu, u64 val);
1554 
1555 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1556 static u64 __kvm_read_sanitised_id_reg(const struct kvm_vcpu *vcpu,
1557 				       const struct sys_reg_desc *r)
1558 {
1559 	u32 id = reg_to_encoding(r);
1560 	u64 val;
1561 
1562 	if (sysreg_visible_as_raz(vcpu, r))
1563 		return 0;
1564 
1565 	val = read_sanitised_ftr_reg(id);
1566 
1567 	switch (id) {
1568 	case SYS_ID_AA64DFR0_EL1:
1569 		val = sanitise_id_aa64dfr0_el1(vcpu, val);
1570 		break;
1571 	case SYS_ID_AA64PFR0_EL1:
1572 		val = sanitise_id_aa64pfr0_el1(vcpu, val);
1573 		break;
1574 	case SYS_ID_AA64PFR1_EL1:
1575 		if (!kvm_has_mte(vcpu->kvm))
1576 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTE);
1577 
1578 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_SME);
1579 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_RNDR_trap);
1580 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_NMI);
1581 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTE_frac);
1582 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_GCS);
1583 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_THE);
1584 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTEX);
1585 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_DF2);
1586 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_PFAR);
1587 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MPAM_frac);
1588 		break;
1589 	case SYS_ID_AA64PFR2_EL1:
1590 		/* We only expose FPMR */
1591 		val &= ID_AA64PFR2_EL1_FPMR;
1592 		break;
1593 	case SYS_ID_AA64ISAR1_EL1:
1594 		if (!vcpu_has_ptrauth(vcpu))
1595 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_APA) |
1596 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_API) |
1597 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_GPA) |
1598 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_GPI));
1599 		break;
1600 	case SYS_ID_AA64ISAR2_EL1:
1601 		if (!vcpu_has_ptrauth(vcpu))
1602 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_APA3) |
1603 				 ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_GPA3));
1604 		if (!cpus_have_final_cap(ARM64_HAS_WFXT))
1605 			val &= ~ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_WFxT);
1606 		break;
1607 	case SYS_ID_AA64MMFR2_EL1:
1608 		val &= ~ID_AA64MMFR2_EL1_CCIDX_MASK;
1609 		break;
1610 	case SYS_ID_AA64MMFR3_EL1:
1611 		val &= ID_AA64MMFR3_EL1_TCRX | ID_AA64MMFR3_EL1_S1POE |
1612 			ID_AA64MMFR3_EL1_S1PIE;
1613 		break;
1614 	case SYS_ID_MMFR4_EL1:
1615 		val &= ~ARM64_FEATURE_MASK(ID_MMFR4_EL1_CCIDX);
1616 		break;
1617 	}
1618 
1619 	return val;
1620 }
1621 
1622 static u64 kvm_read_sanitised_id_reg(struct kvm_vcpu *vcpu,
1623 				     const struct sys_reg_desc *r)
1624 {
1625 	return __kvm_read_sanitised_id_reg(vcpu, r);
1626 }
1627 
1628 static u64 read_id_reg(const struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1629 {
1630 	return kvm_read_vm_id_reg(vcpu->kvm, reg_to_encoding(r));
1631 }
1632 
1633 static bool is_feature_id_reg(u32 encoding)
1634 {
1635 	return (sys_reg_Op0(encoding) == 3 &&
1636 		(sys_reg_Op1(encoding) < 2 || sys_reg_Op1(encoding) == 3) &&
1637 		sys_reg_CRn(encoding) == 0 &&
1638 		sys_reg_CRm(encoding) <= 7);
1639 }
1640 
1641 /*
1642  * Return true if the register's (Op0, Op1, CRn, CRm, Op2) is
1643  * (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8, which is the range of ID
1644  * registers KVM maintains on a per-VM basis.
1645  */
1646 static inline bool is_vm_ftr_id_reg(u32 id)
1647 {
1648 	if (id == SYS_CTR_EL0)
1649 		return true;
1650 
1651 	return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1652 		sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1653 		sys_reg_CRm(id) < 8);
1654 }
1655 
1656 static inline bool is_vcpu_ftr_id_reg(u32 id)
1657 {
1658 	return is_feature_id_reg(id) && !is_vm_ftr_id_reg(id);
1659 }
1660 
1661 static inline bool is_aa32_id_reg(u32 id)
1662 {
1663 	return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1664 		sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1665 		sys_reg_CRm(id) <= 3);
1666 }
1667 
1668 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1669 				  const struct sys_reg_desc *r)
1670 {
1671 	u32 id = reg_to_encoding(r);
1672 
1673 	switch (id) {
1674 	case SYS_ID_AA64ZFR0_EL1:
1675 		if (!vcpu_has_sve(vcpu))
1676 			return REG_RAZ;
1677 		break;
1678 	}
1679 
1680 	return 0;
1681 }
1682 
1683 static unsigned int aa32_id_visibility(const struct kvm_vcpu *vcpu,
1684 				       const struct sys_reg_desc *r)
1685 {
1686 	/*
1687 	 * AArch32 ID registers are UNKNOWN if AArch32 isn't implemented at any
1688 	 * EL. Promote to RAZ/WI in order to guarantee consistency between
1689 	 * systems.
1690 	 */
1691 	if (!kvm_supports_32bit_el0())
1692 		return REG_RAZ | REG_USER_WI;
1693 
1694 	return id_visibility(vcpu, r);
1695 }
1696 
1697 static unsigned int raz_visibility(const struct kvm_vcpu *vcpu,
1698 				   const struct sys_reg_desc *r)
1699 {
1700 	return REG_RAZ;
1701 }
1702 
1703 /* cpufeature ID register access trap handlers */
1704 
1705 static bool access_id_reg(struct kvm_vcpu *vcpu,
1706 			  struct sys_reg_params *p,
1707 			  const struct sys_reg_desc *r)
1708 {
1709 	if (p->is_write)
1710 		return write_to_read_only(vcpu, p, r);
1711 
1712 	p->regval = read_id_reg(vcpu, r);
1713 
1714 	return true;
1715 }
1716 
1717 /* Visibility overrides for SVE-specific control registers */
1718 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1719 				   const struct sys_reg_desc *rd)
1720 {
1721 	if (vcpu_has_sve(vcpu))
1722 		return 0;
1723 
1724 	return REG_HIDDEN;
1725 }
1726 
1727 static unsigned int sme_visibility(const struct kvm_vcpu *vcpu,
1728 				   const struct sys_reg_desc *rd)
1729 {
1730 	if (kvm_has_feat(vcpu->kvm, ID_AA64PFR1_EL1, SME, IMP))
1731 		return 0;
1732 
1733 	return REG_HIDDEN;
1734 }
1735 
1736 static unsigned int fp8_visibility(const struct kvm_vcpu *vcpu,
1737 				   const struct sys_reg_desc *rd)
1738 {
1739 	if (kvm_has_fpmr(vcpu->kvm))
1740 		return 0;
1741 
1742 	return REG_HIDDEN;
1743 }
1744 
1745 static u64 sanitise_id_aa64pfr0_el1(const struct kvm_vcpu *vcpu, u64 val)
1746 {
1747 	if (!vcpu_has_sve(vcpu))
1748 		val &= ~ID_AA64PFR0_EL1_SVE_MASK;
1749 
1750 	/*
1751 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
1752 	 * Although this is a per-CPU feature, we make it global because
1753 	 * asymmetric systems are just a nuisance.
1754 	 *
1755 	 * Userspace can override this as long as it doesn't promise
1756 	 * the impossible.
1757 	 */
1758 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) {
1759 		val &= ~ID_AA64PFR0_EL1_CSV2_MASK;
1760 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV2, IMP);
1761 	}
1762 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) {
1763 		val &= ~ID_AA64PFR0_EL1_CSV3_MASK;
1764 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV3, IMP);
1765 	}
1766 
1767 	if (kvm_vgic_global_state.type == VGIC_V3) {
1768 		val &= ~ID_AA64PFR0_EL1_GIC_MASK;
1769 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, GIC, IMP);
1770 	}
1771 
1772 	val &= ~ID_AA64PFR0_EL1_AMU_MASK;
1773 
1774 	/*
1775 	 * MPAM is disabled by default as KVM also needs a set of PARTID to
1776 	 * program the MPAMVPMx_EL2 PARTID remapping registers with. But some
1777 	 * older kernels let the guest see the ID bit.
1778 	 */
1779 	val &= ~ID_AA64PFR0_EL1_MPAM_MASK;
1780 
1781 	return val;
1782 }
1783 
1784 #define ID_REG_LIMIT_FIELD_ENUM(val, reg, field, limit)			       \
1785 ({									       \
1786 	u64 __f_val = FIELD_GET(reg##_##field##_MASK, val);		       \
1787 	(val) &= ~reg##_##field##_MASK;					       \
1788 	(val) |= FIELD_PREP(reg##_##field##_MASK,			       \
1789 			    min(__f_val,				       \
1790 				(u64)SYS_FIELD_VALUE(reg, field, limit)));     \
1791 	(val);								       \
1792 })
1793 
1794 static u64 sanitise_id_aa64dfr0_el1(const struct kvm_vcpu *vcpu, u64 val)
1795 {
1796 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_AA64DFR0_EL1, DebugVer, V8P8);
1797 
1798 	/*
1799 	 * Only initialize the PMU version if the vCPU was configured with one.
1800 	 */
1801 	val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
1802 	if (kvm_vcpu_has_pmu(vcpu))
1803 		val |= SYS_FIELD_PREP(ID_AA64DFR0_EL1, PMUVer,
1804 				      kvm_arm_pmu_get_pmuver_limit());
1805 
1806 	/* Hide SPE from guests */
1807 	val &= ~ID_AA64DFR0_EL1_PMSVer_MASK;
1808 
1809 	return val;
1810 }
1811 
1812 static int set_id_aa64dfr0_el1(struct kvm_vcpu *vcpu,
1813 			       const struct sys_reg_desc *rd,
1814 			       u64 val)
1815 {
1816 	u8 debugver = SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, val);
1817 	u8 pmuver = SYS_FIELD_GET(ID_AA64DFR0_EL1, PMUVer, val);
1818 
1819 	/*
1820 	 * Prior to commit 3d0dba5764b9 ("KVM: arm64: PMU: Move the
1821 	 * ID_AA64DFR0_EL1.PMUver limit to VM creation"), KVM erroneously
1822 	 * exposed an IMP_DEF PMU to userspace and the guest on systems w/
1823 	 * non-architectural PMUs. Of course, PMUv3 is the only game in town for
1824 	 * PMU virtualization, so the IMP_DEF value was rather user-hostile.
1825 	 *
1826 	 * At minimum, we're on the hook to allow values that were given to
1827 	 * userspace by KVM. Cover our tracks here and replace the IMP_DEF value
1828 	 * with a more sensible NI. The value of an ID register changing under
1829 	 * the nose of the guest is unfortunate, but is certainly no more
1830 	 * surprising than an ill-guided PMU driver poking at impdef system
1831 	 * registers that end in an UNDEF...
1832 	 */
1833 	if (pmuver == ID_AA64DFR0_EL1_PMUVer_IMP_DEF)
1834 		val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
1835 
1836 	/*
1837 	 * ID_AA64DFR0_EL1.DebugVer is one of those awkward fields with a
1838 	 * nonzero minimum safe value.
1839 	 */
1840 	if (debugver < ID_AA64DFR0_EL1_DebugVer_IMP)
1841 		return -EINVAL;
1842 
1843 	return set_id_reg(vcpu, rd, val);
1844 }
1845 
1846 static u64 read_sanitised_id_dfr0_el1(struct kvm_vcpu *vcpu,
1847 				      const struct sys_reg_desc *rd)
1848 {
1849 	u8 perfmon = pmuver_to_perfmon(kvm_arm_pmu_get_pmuver_limit());
1850 	u64 val = read_sanitised_ftr_reg(SYS_ID_DFR0_EL1);
1851 
1852 	val &= ~ID_DFR0_EL1_PerfMon_MASK;
1853 	if (kvm_vcpu_has_pmu(vcpu))
1854 		val |= SYS_FIELD_PREP(ID_DFR0_EL1, PerfMon, perfmon);
1855 
1856 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_DFR0_EL1, CopDbg, Debugv8p8);
1857 
1858 	return val;
1859 }
1860 
1861 static int set_id_dfr0_el1(struct kvm_vcpu *vcpu,
1862 			   const struct sys_reg_desc *rd,
1863 			   u64 val)
1864 {
1865 	u8 perfmon = SYS_FIELD_GET(ID_DFR0_EL1, PerfMon, val);
1866 	u8 copdbg = SYS_FIELD_GET(ID_DFR0_EL1, CopDbg, val);
1867 
1868 	if (perfmon == ID_DFR0_EL1_PerfMon_IMPDEF) {
1869 		val &= ~ID_DFR0_EL1_PerfMon_MASK;
1870 		perfmon = 0;
1871 	}
1872 
1873 	/*
1874 	 * Allow DFR0_EL1.PerfMon to be set from userspace as long as
1875 	 * it doesn't promise more than what the HW gives us on the
1876 	 * AArch64 side (as everything is emulated with that), and
1877 	 * that this is a PMUv3.
1878 	 */
1879 	if (perfmon != 0 && perfmon < ID_DFR0_EL1_PerfMon_PMUv3)
1880 		return -EINVAL;
1881 
1882 	if (copdbg < ID_DFR0_EL1_CopDbg_Armv8)
1883 		return -EINVAL;
1884 
1885 	return set_id_reg(vcpu, rd, val);
1886 }
1887 
1888 static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
1889 			       const struct sys_reg_desc *rd, u64 user_val)
1890 {
1891 	u64 hw_val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1892 	u64 mpam_mask = ID_AA64PFR0_EL1_MPAM_MASK;
1893 
1894 	/*
1895 	 * Commit 011e5f5bf529f ("arm64/cpufeature: Add remaining feature bits
1896 	 * in ID_AA64PFR0 register") exposed the MPAM field of AA64PFR0_EL1 to
1897 	 * guests, but didn't add trap handling. KVM doesn't support MPAM and
1898 	 * always returns an UNDEF for these registers. The guest must see 0
1899 	 * for this field.
1900 	 *
1901 	 * But KVM must also accept values from user-space that were provided
1902 	 * by KVM. On CPUs that support MPAM, permit user-space to write
1903 	 * the sanitizied value to ID_AA64PFR0_EL1.MPAM, but ignore this field.
1904 	 */
1905 	if ((hw_val & mpam_mask) == (user_val & mpam_mask))
1906 		user_val &= ~ID_AA64PFR0_EL1_MPAM_MASK;
1907 
1908 	return set_id_reg(vcpu, rd, user_val);
1909 }
1910 
1911 static int set_id_aa64pfr1_el1(struct kvm_vcpu *vcpu,
1912 			       const struct sys_reg_desc *rd, u64 user_val)
1913 {
1914 	u64 hw_val = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1915 	u64 mpam_mask = ID_AA64PFR1_EL1_MPAM_frac_MASK;
1916 
1917 	/* See set_id_aa64pfr0_el1 for comment about MPAM */
1918 	if ((hw_val & mpam_mask) == (user_val & mpam_mask))
1919 		user_val &= ~ID_AA64PFR1_EL1_MPAM_frac_MASK;
1920 
1921 	return set_id_reg(vcpu, rd, user_val);
1922 }
1923 
1924 static int set_ctr_el0(struct kvm_vcpu *vcpu,
1925 		       const struct sys_reg_desc *rd, u64 user_val)
1926 {
1927 	u8 user_L1Ip = SYS_FIELD_GET(CTR_EL0, L1Ip, user_val);
1928 
1929 	/*
1930 	 * Both AIVIVT (0b01) and VPIPT (0b00) are documented as reserved.
1931 	 * Hence only allow to set VIPT(0b10) or PIPT(0b11) for L1Ip based
1932 	 * on what hardware reports.
1933 	 *
1934 	 * Using a VIPT software model on PIPT will lead to over invalidation,
1935 	 * but still correct. Hence, we can allow downgrading PIPT to VIPT,
1936 	 * but not the other way around. This is handled via arm64_ftr_safe_value()
1937 	 * as CTR_EL0 ftr_bits has L1Ip field with type FTR_EXACT and safe value
1938 	 * set as VIPT.
1939 	 */
1940 	switch (user_L1Ip) {
1941 	case CTR_EL0_L1Ip_RESERVED_VPIPT:
1942 	case CTR_EL0_L1Ip_RESERVED_AIVIVT:
1943 		return -EINVAL;
1944 	case CTR_EL0_L1Ip_VIPT:
1945 	case CTR_EL0_L1Ip_PIPT:
1946 		return set_id_reg(vcpu, rd, user_val);
1947 	default:
1948 		return -ENOENT;
1949 	}
1950 }
1951 
1952 /*
1953  * cpufeature ID register user accessors
1954  *
1955  * For now, these registers are immutable for userspace, so no values
1956  * are stored, and for set_id_reg() we don't allow the effective value
1957  * to be changed.
1958  */
1959 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1960 		      u64 *val)
1961 {
1962 	/*
1963 	 * Avoid locking if the VM has already started, as the ID registers are
1964 	 * guaranteed to be invariant at that point.
1965 	 */
1966 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
1967 		*val = read_id_reg(vcpu, rd);
1968 		return 0;
1969 	}
1970 
1971 	mutex_lock(&vcpu->kvm->arch.config_lock);
1972 	*val = read_id_reg(vcpu, rd);
1973 	mutex_unlock(&vcpu->kvm->arch.config_lock);
1974 
1975 	return 0;
1976 }
1977 
1978 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1979 		      u64 val)
1980 {
1981 	u32 id = reg_to_encoding(rd);
1982 	int ret;
1983 
1984 	mutex_lock(&vcpu->kvm->arch.config_lock);
1985 
1986 	/*
1987 	 * Once the VM has started the ID registers are immutable. Reject any
1988 	 * write that does not match the final register value.
1989 	 */
1990 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
1991 		if (val != read_id_reg(vcpu, rd))
1992 			ret = -EBUSY;
1993 		else
1994 			ret = 0;
1995 
1996 		mutex_unlock(&vcpu->kvm->arch.config_lock);
1997 		return ret;
1998 	}
1999 
2000 	ret = arm64_check_features(vcpu, rd, val);
2001 	if (!ret)
2002 		kvm_set_vm_id_reg(vcpu->kvm, id, val);
2003 
2004 	mutex_unlock(&vcpu->kvm->arch.config_lock);
2005 
2006 	/*
2007 	 * arm64_check_features() returns -E2BIG to indicate the register's
2008 	 * feature set is a superset of the maximally-allowed register value.
2009 	 * While it would be nice to precisely describe this to userspace, the
2010 	 * existing UAPI for KVM_SET_ONE_REG has it that invalid register
2011 	 * writes return -EINVAL.
2012 	 */
2013 	if (ret == -E2BIG)
2014 		ret = -EINVAL;
2015 	return ret;
2016 }
2017 
2018 void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val)
2019 {
2020 	u64 *p = __vm_id_reg(&kvm->arch, reg);
2021 
2022 	lockdep_assert_held(&kvm->arch.config_lock);
2023 
2024 	if (KVM_BUG_ON(kvm_vm_has_ran_once(kvm) || !p, kvm))
2025 		return;
2026 
2027 	*p = val;
2028 }
2029 
2030 static int get_raz_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2031 		       u64 *val)
2032 {
2033 	*val = 0;
2034 	return 0;
2035 }
2036 
2037 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2038 		      u64 val)
2039 {
2040 	return 0;
2041 }
2042 
2043 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2044 		       const struct sys_reg_desc *r)
2045 {
2046 	if (p->is_write)
2047 		return write_to_read_only(vcpu, p, r);
2048 
2049 	p->regval = kvm_read_vm_id_reg(vcpu->kvm, SYS_CTR_EL0);
2050 	return true;
2051 }
2052 
2053 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2054 			 const struct sys_reg_desc *r)
2055 {
2056 	if (p->is_write)
2057 		return write_to_read_only(vcpu, p, r);
2058 
2059 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
2060 	return true;
2061 }
2062 
2063 /*
2064  * Fabricate a CLIDR_EL1 value instead of using the real value, which can vary
2065  * by the physical CPU which the vcpu currently resides in.
2066  */
2067 static u64 reset_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2068 {
2069 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2070 	u64 clidr;
2071 	u8 loc;
2072 
2073 	if ((ctr_el0 & CTR_EL0_IDC)) {
2074 		/*
2075 		 * Data cache clean to the PoU is not required so LoUU and LoUIS
2076 		 * will not be set and a unified cache, which will be marked as
2077 		 * LoC, will be added.
2078 		 *
2079 		 * If not DIC, let the unified cache L2 so that an instruction
2080 		 * cache can be added as L1 later.
2081 		 */
2082 		loc = (ctr_el0 & CTR_EL0_DIC) ? 1 : 2;
2083 		clidr = CACHE_TYPE_UNIFIED << CLIDR_CTYPE_SHIFT(loc);
2084 	} else {
2085 		/*
2086 		 * Data cache clean to the PoU is required so let L1 have a data
2087 		 * cache and mark it as LoUU and LoUIS. As L1 has a data cache,
2088 		 * it can be marked as LoC too.
2089 		 */
2090 		loc = 1;
2091 		clidr = 1 << CLIDR_LOUU_SHIFT;
2092 		clidr |= 1 << CLIDR_LOUIS_SHIFT;
2093 		clidr |= CACHE_TYPE_DATA << CLIDR_CTYPE_SHIFT(1);
2094 	}
2095 
2096 	/*
2097 	 * Instruction cache invalidation to the PoU is required so let L1 have
2098 	 * an instruction cache. If L1 already has a data cache, it will be
2099 	 * CACHE_TYPE_SEPARATE.
2100 	 */
2101 	if (!(ctr_el0 & CTR_EL0_DIC))
2102 		clidr |= CACHE_TYPE_INST << CLIDR_CTYPE_SHIFT(1);
2103 
2104 	clidr |= loc << CLIDR_LOC_SHIFT;
2105 
2106 	/*
2107 	 * Add tag cache unified to data cache. Allocation tags and data are
2108 	 * unified in a cache line so that it looks valid even if there is only
2109 	 * one cache line.
2110 	 */
2111 	if (kvm_has_mte(vcpu->kvm))
2112 		clidr |= 2ULL << CLIDR_TTYPE_SHIFT(loc);
2113 
2114 	__vcpu_sys_reg(vcpu, r->reg) = clidr;
2115 
2116 	return __vcpu_sys_reg(vcpu, r->reg);
2117 }
2118 
2119 static int set_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2120 		      u64 val)
2121 {
2122 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2123 	u64 idc = !CLIDR_LOC(val) || (!CLIDR_LOUIS(val) && !CLIDR_LOUU(val));
2124 
2125 	if ((val & CLIDR_EL1_RES0) || (!(ctr_el0 & CTR_EL0_IDC) && idc))
2126 		return -EINVAL;
2127 
2128 	__vcpu_sys_reg(vcpu, rd->reg) = val;
2129 
2130 	return 0;
2131 }
2132 
2133 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2134 			  const struct sys_reg_desc *r)
2135 {
2136 	int reg = r->reg;
2137 
2138 	if (p->is_write)
2139 		vcpu_write_sys_reg(vcpu, p->regval, reg);
2140 	else
2141 		p->regval = vcpu_read_sys_reg(vcpu, reg);
2142 	return true;
2143 }
2144 
2145 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2146 			  const struct sys_reg_desc *r)
2147 {
2148 	u32 csselr;
2149 
2150 	if (p->is_write)
2151 		return write_to_read_only(vcpu, p, r);
2152 
2153 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
2154 	csselr &= CSSELR_EL1_Level | CSSELR_EL1_InD;
2155 	if (csselr < CSSELR_MAX)
2156 		p->regval = get_ccsidr(vcpu, csselr);
2157 
2158 	return true;
2159 }
2160 
2161 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
2162 				   const struct sys_reg_desc *rd)
2163 {
2164 	if (kvm_has_mte(vcpu->kvm))
2165 		return 0;
2166 
2167 	return REG_HIDDEN;
2168 }
2169 
2170 #define MTE_REG(name) {				\
2171 	SYS_DESC(SYS_##name),			\
2172 	.access = undef_access,			\
2173 	.reset = reset_unknown,			\
2174 	.reg = name,				\
2175 	.visibility = mte_visibility,		\
2176 }
2177 
2178 static unsigned int el2_visibility(const struct kvm_vcpu *vcpu,
2179 				   const struct sys_reg_desc *rd)
2180 {
2181 	if (vcpu_has_nv(vcpu))
2182 		return 0;
2183 
2184 	return REG_HIDDEN;
2185 }
2186 
2187 static bool bad_vncr_trap(struct kvm_vcpu *vcpu,
2188 			  struct sys_reg_params *p,
2189 			  const struct sys_reg_desc *r)
2190 {
2191 	/*
2192 	 * We really shouldn't be here, and this is likely the result
2193 	 * of a misconfigured trap, as this register should target the
2194 	 * VNCR page, and nothing else.
2195 	 */
2196 	return bad_trap(vcpu, p, r,
2197 			"trap of VNCR-backed register");
2198 }
2199 
2200 static bool bad_redir_trap(struct kvm_vcpu *vcpu,
2201 			   struct sys_reg_params *p,
2202 			   const struct sys_reg_desc *r)
2203 {
2204 	/*
2205 	 * We really shouldn't be here, and this is likely the result
2206 	 * of a misconfigured trap, as this register should target the
2207 	 * corresponding EL1, and nothing else.
2208 	 */
2209 	return bad_trap(vcpu, p, r,
2210 			"trap of EL2 register redirected to EL1");
2211 }
2212 
2213 #define EL2_REG(name, acc, rst, v) {		\
2214 	SYS_DESC(SYS_##name),			\
2215 	.access = acc,				\
2216 	.reset = rst,				\
2217 	.reg = name,				\
2218 	.visibility = el2_visibility,		\
2219 	.val = v,				\
2220 }
2221 
2222 #define EL2_REG_FILTERED(name, acc, rst, v, filter) {	\
2223 	SYS_DESC(SYS_##name),			\
2224 	.access = acc,				\
2225 	.reset = rst,				\
2226 	.reg = name,				\
2227 	.visibility = filter,			\
2228 	.val = v,				\
2229 }
2230 
2231 #define EL2_REG_VNCR(name, rst, v)	EL2_REG(name, bad_vncr_trap, rst, v)
2232 #define EL2_REG_REDIR(name, rst, v)	EL2_REG(name, bad_redir_trap, rst, v)
2233 
2234 /*
2235  * Since reset() callback and field val are not used for idregs, they will be
2236  * used for specific purposes for idregs.
2237  * The reset() would return KVM sanitised register value. The value would be the
2238  * same as the host kernel sanitised value if there is no KVM sanitisation.
2239  * The val would be used as a mask indicating writable fields for the idreg.
2240  * Only bits with 1 are writable from userspace. This mask might not be
2241  * necessary in the future whenever all ID registers are enabled as writable
2242  * from userspace.
2243  */
2244 
2245 #define ID_DESC(name)				\
2246 	SYS_DESC(SYS_##name),			\
2247 	.access	= access_id_reg,		\
2248 	.get_user = get_id_reg			\
2249 
2250 /* sys_reg_desc initialiser for known cpufeature ID registers */
2251 #define ID_SANITISED(name) {			\
2252 	ID_DESC(name),				\
2253 	.set_user = set_id_reg,			\
2254 	.visibility = id_visibility,		\
2255 	.reset = kvm_read_sanitised_id_reg,	\
2256 	.val = 0,				\
2257 }
2258 
2259 /* sys_reg_desc initialiser for known cpufeature ID registers */
2260 #define AA32_ID_SANITISED(name) {		\
2261 	ID_DESC(name),				\
2262 	.set_user = set_id_reg,			\
2263 	.visibility = aa32_id_visibility,	\
2264 	.reset = kvm_read_sanitised_id_reg,	\
2265 	.val = 0,				\
2266 }
2267 
2268 /* sys_reg_desc initialiser for writable ID registers */
2269 #define ID_WRITABLE(name, mask) {		\
2270 	ID_DESC(name),				\
2271 	.set_user = set_id_reg,			\
2272 	.visibility = id_visibility,		\
2273 	.reset = kvm_read_sanitised_id_reg,	\
2274 	.val = mask,				\
2275 }
2276 
2277 /* sys_reg_desc initialiser for cpufeature ID registers that need filtering */
2278 #define ID_FILTERED(sysreg, name, mask) {	\
2279 	ID_DESC(sysreg),				\
2280 	.set_user = set_##name,				\
2281 	.visibility = id_visibility,			\
2282 	.reset = kvm_read_sanitised_id_reg,		\
2283 	.val = (mask),					\
2284 }
2285 
2286 /*
2287  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
2288  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
2289  * (1 <= crm < 8, 0 <= Op2 < 8).
2290  */
2291 #define ID_UNALLOCATED(crm, op2) {			\
2292 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
2293 	.access = access_id_reg,			\
2294 	.get_user = get_id_reg,				\
2295 	.set_user = set_id_reg,				\
2296 	.visibility = raz_visibility,			\
2297 	.reset = kvm_read_sanitised_id_reg,		\
2298 	.val = 0,					\
2299 }
2300 
2301 /*
2302  * sys_reg_desc initialiser for known ID registers that we hide from guests.
2303  * For now, these are exposed just like unallocated ID regs: they appear
2304  * RAZ for the guest.
2305  */
2306 #define ID_HIDDEN(name) {			\
2307 	ID_DESC(name),				\
2308 	.set_user = set_id_reg,			\
2309 	.visibility = raz_visibility,		\
2310 	.reset = kvm_read_sanitised_id_reg,	\
2311 	.val = 0,				\
2312 }
2313 
2314 static bool access_sp_el1(struct kvm_vcpu *vcpu,
2315 			  struct sys_reg_params *p,
2316 			  const struct sys_reg_desc *r)
2317 {
2318 	if (p->is_write)
2319 		__vcpu_sys_reg(vcpu, SP_EL1) = p->regval;
2320 	else
2321 		p->regval = __vcpu_sys_reg(vcpu, SP_EL1);
2322 
2323 	return true;
2324 }
2325 
2326 static bool access_elr(struct kvm_vcpu *vcpu,
2327 		       struct sys_reg_params *p,
2328 		       const struct sys_reg_desc *r)
2329 {
2330 	if (p->is_write)
2331 		vcpu_write_sys_reg(vcpu, p->regval, ELR_EL1);
2332 	else
2333 		p->regval = vcpu_read_sys_reg(vcpu, ELR_EL1);
2334 
2335 	return true;
2336 }
2337 
2338 static bool access_spsr(struct kvm_vcpu *vcpu,
2339 			struct sys_reg_params *p,
2340 			const struct sys_reg_desc *r)
2341 {
2342 	if (p->is_write)
2343 		__vcpu_sys_reg(vcpu, SPSR_EL1) = p->regval;
2344 	else
2345 		p->regval = __vcpu_sys_reg(vcpu, SPSR_EL1);
2346 
2347 	return true;
2348 }
2349 
2350 static bool access_cntkctl_el12(struct kvm_vcpu *vcpu,
2351 				struct sys_reg_params *p,
2352 				const struct sys_reg_desc *r)
2353 {
2354 	if (p->is_write)
2355 		__vcpu_sys_reg(vcpu, CNTKCTL_EL1) = p->regval;
2356 	else
2357 		p->regval = __vcpu_sys_reg(vcpu, CNTKCTL_EL1);
2358 
2359 	return true;
2360 }
2361 
2362 static u64 reset_hcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2363 {
2364 	u64 val = r->val;
2365 
2366 	if (!cpus_have_final_cap(ARM64_HAS_HCR_NV1))
2367 		val |= HCR_E2H;
2368 
2369 	return __vcpu_sys_reg(vcpu, r->reg) = val;
2370 }
2371 
2372 static unsigned int __el2_visibility(const struct kvm_vcpu *vcpu,
2373 				     const struct sys_reg_desc *rd,
2374 				     unsigned int (*fn)(const struct kvm_vcpu *,
2375 							const struct sys_reg_desc *))
2376 {
2377 	return el2_visibility(vcpu, rd) ?: fn(vcpu, rd);
2378 }
2379 
2380 static unsigned int sve_el2_visibility(const struct kvm_vcpu *vcpu,
2381 				       const struct sys_reg_desc *rd)
2382 {
2383 	return __el2_visibility(vcpu, rd, sve_visibility);
2384 }
2385 
2386 static bool access_zcr_el2(struct kvm_vcpu *vcpu,
2387 			   struct sys_reg_params *p,
2388 			   const struct sys_reg_desc *r)
2389 {
2390 	unsigned int vq;
2391 
2392 	if (guest_hyp_sve_traps_enabled(vcpu)) {
2393 		kvm_inject_nested_sve_trap(vcpu);
2394 		return true;
2395 	}
2396 
2397 	if (!p->is_write) {
2398 		p->regval = vcpu_read_sys_reg(vcpu, ZCR_EL2);
2399 		return true;
2400 	}
2401 
2402 	vq = SYS_FIELD_GET(ZCR_ELx, LEN, p->regval) + 1;
2403 	vq = min(vq, vcpu_sve_max_vq(vcpu));
2404 	vcpu_write_sys_reg(vcpu, vq - 1, ZCR_EL2);
2405 	return true;
2406 }
2407 
2408 static unsigned int s1poe_visibility(const struct kvm_vcpu *vcpu,
2409 				     const struct sys_reg_desc *rd)
2410 {
2411 	if (kvm_has_s1poe(vcpu->kvm))
2412 		return 0;
2413 
2414 	return REG_HIDDEN;
2415 }
2416 
2417 static unsigned int s1poe_el2_visibility(const struct kvm_vcpu *vcpu,
2418 					 const struct sys_reg_desc *rd)
2419 {
2420 	return __el2_visibility(vcpu, rd, s1poe_visibility);
2421 }
2422 
2423 static unsigned int tcr2_visibility(const struct kvm_vcpu *vcpu,
2424 				    const struct sys_reg_desc *rd)
2425 {
2426 	if (kvm_has_tcr2(vcpu->kvm))
2427 		return 0;
2428 
2429 	return REG_HIDDEN;
2430 }
2431 
2432 static unsigned int tcr2_el2_visibility(const struct kvm_vcpu *vcpu,
2433 				    const struct sys_reg_desc *rd)
2434 {
2435 	return __el2_visibility(vcpu, rd, tcr2_visibility);
2436 }
2437 
2438 static unsigned int s1pie_visibility(const struct kvm_vcpu *vcpu,
2439 				     const struct sys_reg_desc *rd)
2440 {
2441 	if (kvm_has_s1pie(vcpu->kvm))
2442 		return 0;
2443 
2444 	return REG_HIDDEN;
2445 }
2446 
2447 static unsigned int s1pie_el2_visibility(const struct kvm_vcpu *vcpu,
2448 					 const struct sys_reg_desc *rd)
2449 {
2450 	return __el2_visibility(vcpu, rd, s1pie_visibility);
2451 }
2452 
2453 /*
2454  * Architected system registers.
2455  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
2456  *
2457  * Debug handling: We do trap most, if not all debug related system
2458  * registers. The implementation is good enough to ensure that a guest
2459  * can use these with minimal performance degradation. The drawback is
2460  * that we don't implement any of the external debug architecture.
2461  * This should be revisited if we ever encounter a more demanding
2462  * guest...
2463  */
2464 static const struct sys_reg_desc sys_reg_descs[] = {
2465 	DBG_BCR_BVR_WCR_WVR_EL1(0),
2466 	DBG_BCR_BVR_WCR_WVR_EL1(1),
2467 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
2468 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
2469 	DBG_BCR_BVR_WCR_WVR_EL1(2),
2470 	DBG_BCR_BVR_WCR_WVR_EL1(3),
2471 	DBG_BCR_BVR_WCR_WVR_EL1(4),
2472 	DBG_BCR_BVR_WCR_WVR_EL1(5),
2473 	DBG_BCR_BVR_WCR_WVR_EL1(6),
2474 	DBG_BCR_BVR_WCR_WVR_EL1(7),
2475 	DBG_BCR_BVR_WCR_WVR_EL1(8),
2476 	DBG_BCR_BVR_WCR_WVR_EL1(9),
2477 	DBG_BCR_BVR_WCR_WVR_EL1(10),
2478 	DBG_BCR_BVR_WCR_WVR_EL1(11),
2479 	DBG_BCR_BVR_WCR_WVR_EL1(12),
2480 	DBG_BCR_BVR_WCR_WVR_EL1(13),
2481 	DBG_BCR_BVR_WCR_WVR_EL1(14),
2482 	DBG_BCR_BVR_WCR_WVR_EL1(15),
2483 
2484 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
2485 	{ SYS_DESC(SYS_OSLAR_EL1), trap_oslar_el1 },
2486 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1, reset_val, OSLSR_EL1,
2487 		OSLSR_EL1_OSLM_IMPLEMENTED, .set_user = set_oslsr_el1, },
2488 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
2489 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
2490 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
2491 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
2492 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
2493 
2494 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
2495 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
2496 	// DBGDTR[TR]X_EL0 share the same encoding
2497 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
2498 
2499 	{ SYS_DESC(SYS_DBGVCR32_EL2), undef_access, reset_val, DBGVCR32_EL2, 0 },
2500 
2501 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
2502 
2503 	/*
2504 	 * ID regs: all ID_SANITISED() entries here must have corresponding
2505 	 * entries in arm64_ftr_regs[].
2506 	 */
2507 
2508 	/* AArch64 mappings of the AArch32 ID registers */
2509 	/* CRm=1 */
2510 	AA32_ID_SANITISED(ID_PFR0_EL1),
2511 	AA32_ID_SANITISED(ID_PFR1_EL1),
2512 	{ SYS_DESC(SYS_ID_DFR0_EL1),
2513 	  .access = access_id_reg,
2514 	  .get_user = get_id_reg,
2515 	  .set_user = set_id_dfr0_el1,
2516 	  .visibility = aa32_id_visibility,
2517 	  .reset = read_sanitised_id_dfr0_el1,
2518 	  .val = ID_DFR0_EL1_PerfMon_MASK |
2519 		 ID_DFR0_EL1_CopDbg_MASK, },
2520 	ID_HIDDEN(ID_AFR0_EL1),
2521 	AA32_ID_SANITISED(ID_MMFR0_EL1),
2522 	AA32_ID_SANITISED(ID_MMFR1_EL1),
2523 	AA32_ID_SANITISED(ID_MMFR2_EL1),
2524 	AA32_ID_SANITISED(ID_MMFR3_EL1),
2525 
2526 	/* CRm=2 */
2527 	AA32_ID_SANITISED(ID_ISAR0_EL1),
2528 	AA32_ID_SANITISED(ID_ISAR1_EL1),
2529 	AA32_ID_SANITISED(ID_ISAR2_EL1),
2530 	AA32_ID_SANITISED(ID_ISAR3_EL1),
2531 	AA32_ID_SANITISED(ID_ISAR4_EL1),
2532 	AA32_ID_SANITISED(ID_ISAR5_EL1),
2533 	AA32_ID_SANITISED(ID_MMFR4_EL1),
2534 	AA32_ID_SANITISED(ID_ISAR6_EL1),
2535 
2536 	/* CRm=3 */
2537 	AA32_ID_SANITISED(MVFR0_EL1),
2538 	AA32_ID_SANITISED(MVFR1_EL1),
2539 	AA32_ID_SANITISED(MVFR2_EL1),
2540 	ID_UNALLOCATED(3,3),
2541 	AA32_ID_SANITISED(ID_PFR2_EL1),
2542 	ID_HIDDEN(ID_DFR1_EL1),
2543 	AA32_ID_SANITISED(ID_MMFR5_EL1),
2544 	ID_UNALLOCATED(3,7),
2545 
2546 	/* AArch64 ID registers */
2547 	/* CRm=4 */
2548 	ID_FILTERED(ID_AA64PFR0_EL1, id_aa64pfr0_el1,
2549 		    ~(ID_AA64PFR0_EL1_AMU |
2550 		      ID_AA64PFR0_EL1_MPAM |
2551 		      ID_AA64PFR0_EL1_SVE |
2552 		      ID_AA64PFR0_EL1_RAS |
2553 		      ID_AA64PFR0_EL1_AdvSIMD |
2554 		      ID_AA64PFR0_EL1_FP)),
2555 	ID_FILTERED(ID_AA64PFR1_EL1, id_aa64pfr1_el1,
2556 				     ~(ID_AA64PFR1_EL1_PFAR |
2557 				       ID_AA64PFR1_EL1_DF2 |
2558 				       ID_AA64PFR1_EL1_MTEX |
2559 				       ID_AA64PFR1_EL1_THE |
2560 				       ID_AA64PFR1_EL1_GCS |
2561 				       ID_AA64PFR1_EL1_MTE_frac |
2562 				       ID_AA64PFR1_EL1_NMI |
2563 				       ID_AA64PFR1_EL1_RNDR_trap |
2564 				       ID_AA64PFR1_EL1_SME |
2565 				       ID_AA64PFR1_EL1_RES0 |
2566 				       ID_AA64PFR1_EL1_MPAM_frac |
2567 				       ID_AA64PFR1_EL1_RAS_frac |
2568 				       ID_AA64PFR1_EL1_MTE)),
2569 	ID_WRITABLE(ID_AA64PFR2_EL1, ID_AA64PFR2_EL1_FPMR),
2570 	ID_UNALLOCATED(4,3),
2571 	ID_WRITABLE(ID_AA64ZFR0_EL1, ~ID_AA64ZFR0_EL1_RES0),
2572 	ID_HIDDEN(ID_AA64SMFR0_EL1),
2573 	ID_UNALLOCATED(4,6),
2574 	ID_WRITABLE(ID_AA64FPFR0_EL1, ~ID_AA64FPFR0_EL1_RES0),
2575 
2576 	/* CRm=5 */
2577 	/*
2578 	 * Prior to FEAT_Debugv8.9, the architecture defines context-aware
2579 	 * breakpoints (CTX_CMPs) as the highest numbered breakpoints (BRPs).
2580 	 * KVM does not trap + emulate the breakpoint registers, and as such
2581 	 * cannot support a layout that misaligns with the underlying hardware.
2582 	 * While it may be possible to describe a subset that aligns with
2583 	 * hardware, just prevent changes to BRPs and CTX_CMPs altogether for
2584 	 * simplicity.
2585 	 *
2586 	 * See DDI0487K.a, section D2.8.3 Breakpoint types and linking
2587 	 * of breakpoints for more details.
2588 	 */
2589 	ID_FILTERED(ID_AA64DFR0_EL1, id_aa64dfr0_el1,
2590 		    ID_AA64DFR0_EL1_DoubleLock_MASK |
2591 		    ID_AA64DFR0_EL1_WRPs_MASK |
2592 		    ID_AA64DFR0_EL1_PMUVer_MASK |
2593 		    ID_AA64DFR0_EL1_DebugVer_MASK),
2594 	ID_SANITISED(ID_AA64DFR1_EL1),
2595 	ID_UNALLOCATED(5,2),
2596 	ID_UNALLOCATED(5,3),
2597 	ID_HIDDEN(ID_AA64AFR0_EL1),
2598 	ID_HIDDEN(ID_AA64AFR1_EL1),
2599 	ID_UNALLOCATED(5,6),
2600 	ID_UNALLOCATED(5,7),
2601 
2602 	/* CRm=6 */
2603 	ID_WRITABLE(ID_AA64ISAR0_EL1, ~ID_AA64ISAR0_EL1_RES0),
2604 	ID_WRITABLE(ID_AA64ISAR1_EL1, ~(ID_AA64ISAR1_EL1_GPI |
2605 					ID_AA64ISAR1_EL1_GPA |
2606 					ID_AA64ISAR1_EL1_API |
2607 					ID_AA64ISAR1_EL1_APA)),
2608 	ID_WRITABLE(ID_AA64ISAR2_EL1, ~(ID_AA64ISAR2_EL1_RES0 |
2609 					ID_AA64ISAR2_EL1_APA3 |
2610 					ID_AA64ISAR2_EL1_GPA3)),
2611 	ID_UNALLOCATED(6,3),
2612 	ID_UNALLOCATED(6,4),
2613 	ID_UNALLOCATED(6,5),
2614 	ID_UNALLOCATED(6,6),
2615 	ID_UNALLOCATED(6,7),
2616 
2617 	/* CRm=7 */
2618 	ID_WRITABLE(ID_AA64MMFR0_EL1, ~(ID_AA64MMFR0_EL1_RES0 |
2619 					ID_AA64MMFR0_EL1_TGRAN4_2 |
2620 					ID_AA64MMFR0_EL1_TGRAN64_2 |
2621 					ID_AA64MMFR0_EL1_TGRAN16_2)),
2622 	ID_WRITABLE(ID_AA64MMFR1_EL1, ~(ID_AA64MMFR1_EL1_RES0 |
2623 					ID_AA64MMFR1_EL1_HCX |
2624 					ID_AA64MMFR1_EL1_TWED |
2625 					ID_AA64MMFR1_EL1_XNX |
2626 					ID_AA64MMFR1_EL1_VH |
2627 					ID_AA64MMFR1_EL1_VMIDBits)),
2628 	ID_WRITABLE(ID_AA64MMFR2_EL1, ~(ID_AA64MMFR2_EL1_RES0 |
2629 					ID_AA64MMFR2_EL1_EVT |
2630 					ID_AA64MMFR2_EL1_FWB |
2631 					ID_AA64MMFR2_EL1_IDS |
2632 					ID_AA64MMFR2_EL1_NV |
2633 					ID_AA64MMFR2_EL1_CCIDX)),
2634 	ID_WRITABLE(ID_AA64MMFR3_EL1, (ID_AA64MMFR3_EL1_TCRX	|
2635 				       ID_AA64MMFR3_EL1_S1PIE   |
2636 				       ID_AA64MMFR3_EL1_S1POE)),
2637 	ID_SANITISED(ID_AA64MMFR4_EL1),
2638 	ID_UNALLOCATED(7,5),
2639 	ID_UNALLOCATED(7,6),
2640 	ID_UNALLOCATED(7,7),
2641 
2642 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
2643 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
2644 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
2645 
2646 	MTE_REG(RGSR_EL1),
2647 	MTE_REG(GCR_EL1),
2648 
2649 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
2650 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
2651 	{ SYS_DESC(SYS_SMPRI_EL1), undef_access },
2652 	{ SYS_DESC(SYS_SMCR_EL1), undef_access },
2653 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
2654 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
2655 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
2656 	{ SYS_DESC(SYS_TCR2_EL1), access_vm_reg, reset_val, TCR2_EL1, 0,
2657 	  .visibility = tcr2_visibility },
2658 
2659 	PTRAUTH_KEY(APIA),
2660 	PTRAUTH_KEY(APIB),
2661 	PTRAUTH_KEY(APDA),
2662 	PTRAUTH_KEY(APDB),
2663 	PTRAUTH_KEY(APGA),
2664 
2665 	{ SYS_DESC(SYS_SPSR_EL1), access_spsr},
2666 	{ SYS_DESC(SYS_ELR_EL1), access_elr},
2667 
2668 	{ SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
2669 
2670 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
2671 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
2672 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
2673 
2674 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
2675 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
2676 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
2677 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
2678 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
2679 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
2680 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
2681 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
2682 
2683 	MTE_REG(TFSR_EL1),
2684 	MTE_REG(TFSRE0_EL1),
2685 
2686 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
2687 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
2688 
2689 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
2690 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
2691 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
2692 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
2693 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
2694 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
2695 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
2696 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
2697 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
2698 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
2699 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
2700 	/* PMBIDR_EL1 is not trapped */
2701 
2702 	{ PMU_SYS_REG(PMINTENSET_EL1),
2703 	  .access = access_pminten, .reg = PMINTENSET_EL1,
2704 	  .get_user = get_pmreg, .set_user = set_pmreg },
2705 	{ PMU_SYS_REG(PMINTENCLR_EL1),
2706 	  .access = access_pminten, .reg = PMINTENSET_EL1,
2707 	  .get_user = get_pmreg, .set_user = set_pmreg },
2708 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
2709 
2710 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
2711 	{ SYS_DESC(SYS_PIRE0_EL1), NULL, reset_unknown, PIRE0_EL1,
2712 	  .visibility = s1pie_visibility },
2713 	{ SYS_DESC(SYS_PIR_EL1), NULL, reset_unknown, PIR_EL1,
2714 	  .visibility = s1pie_visibility },
2715 	{ SYS_DESC(SYS_POR_EL1), NULL, reset_unknown, POR_EL1,
2716 	  .visibility = s1poe_visibility },
2717 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
2718 
2719 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
2720 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
2721 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
2722 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
2723 	{ SYS_DESC(SYS_MPAMIDR_EL1), undef_access },
2724 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
2725 
2726 	{ SYS_DESC(SYS_MPAM1_EL1), undef_access },
2727 	{ SYS_DESC(SYS_MPAM0_EL1), undef_access },
2728 	{ SYS_DESC(SYS_VBAR_EL1), access_rw, reset_val, VBAR_EL1, 0 },
2729 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
2730 
2731 	{ SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
2732 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
2733 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
2734 	{ SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
2735 	{ SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
2736 	{ SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
2737 	{ SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
2738 	{ SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
2739 	{ SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
2740 	{ SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
2741 	{ SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
2742 	{ SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
2743 	{ SYS_DESC(SYS_ICC_DIR_EL1), undef_access },
2744 	{ SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
2745 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
2746 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
2747 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
2748 	{ SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
2749 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
2750 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
2751 	{ SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
2752 	{ SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
2753 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
2754 	{ SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
2755 	{ SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
2756 
2757 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
2758 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
2759 
2760 	{ SYS_DESC(SYS_ACCDATA_EL1), undef_access },
2761 
2762 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
2763 
2764 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
2765 
2766 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
2767 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr, reset_clidr, CLIDR_EL1,
2768 	  .set_user = set_clidr, .val = ~CLIDR_EL1_RES0 },
2769 	{ SYS_DESC(SYS_CCSIDR2_EL1), undef_access },
2770 	{ SYS_DESC(SYS_SMIDR_EL1), undef_access },
2771 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
2772 	ID_FILTERED(CTR_EL0, ctr_el0,
2773 		    CTR_EL0_DIC_MASK |
2774 		    CTR_EL0_IDC_MASK |
2775 		    CTR_EL0_DminLine_MASK |
2776 		    CTR_EL0_L1Ip_MASK |
2777 		    CTR_EL0_IminLine_MASK),
2778 	{ SYS_DESC(SYS_SVCR), undef_access, reset_val, SVCR, 0, .visibility = sme_visibility  },
2779 	{ SYS_DESC(SYS_FPMR), undef_access, reset_val, FPMR, 0, .visibility = fp8_visibility },
2780 
2781 	{ PMU_SYS_REG(PMCR_EL0), .access = access_pmcr, .reset = reset_pmcr,
2782 	  .reg = PMCR_EL0, .get_user = get_pmcr, .set_user = set_pmcr },
2783 	{ PMU_SYS_REG(PMCNTENSET_EL0),
2784 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
2785 	  .get_user = get_pmreg, .set_user = set_pmreg },
2786 	{ PMU_SYS_REG(PMCNTENCLR_EL0),
2787 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
2788 	  .get_user = get_pmreg, .set_user = set_pmreg },
2789 	{ PMU_SYS_REG(PMOVSCLR_EL0),
2790 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
2791 	  .get_user = get_pmreg, .set_user = set_pmreg },
2792 	/*
2793 	 * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
2794 	 * previously (and pointlessly) advertised in the past...
2795 	 */
2796 	{ PMU_SYS_REG(PMSWINC_EL0),
2797 	  .get_user = get_raz_reg, .set_user = set_wi_reg,
2798 	  .access = access_pmswinc, .reset = NULL },
2799 	{ PMU_SYS_REG(PMSELR_EL0),
2800 	  .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
2801 	{ PMU_SYS_REG(PMCEID0_EL0),
2802 	  .access = access_pmceid, .reset = NULL },
2803 	{ PMU_SYS_REG(PMCEID1_EL0),
2804 	  .access = access_pmceid, .reset = NULL },
2805 	{ PMU_SYS_REG(PMCCNTR_EL0),
2806 	  .access = access_pmu_evcntr, .reset = reset_unknown,
2807 	  .reg = PMCCNTR_EL0, .get_user = get_pmu_evcntr},
2808 	{ PMU_SYS_REG(PMXEVTYPER_EL0),
2809 	  .access = access_pmu_evtyper, .reset = NULL },
2810 	{ PMU_SYS_REG(PMXEVCNTR_EL0),
2811 	  .access = access_pmu_evcntr, .reset = NULL },
2812 	/*
2813 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
2814 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
2815 	 */
2816 	{ PMU_SYS_REG(PMUSERENR_EL0), .access = access_pmuserenr,
2817 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
2818 	{ PMU_SYS_REG(PMOVSSET_EL0),
2819 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
2820 	  .get_user = get_pmreg, .set_user = set_pmreg },
2821 
2822 	{ SYS_DESC(SYS_POR_EL0), NULL, reset_unknown, POR_EL0,
2823 	  .visibility = s1poe_visibility },
2824 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
2825 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
2826 	{ SYS_DESC(SYS_TPIDR2_EL0), undef_access },
2827 
2828 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
2829 
2830 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
2831 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
2832 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
2833 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
2834 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
2835 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
2836 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
2837 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
2838 	AMU_AMEVCNTR0_EL0(0),
2839 	AMU_AMEVCNTR0_EL0(1),
2840 	AMU_AMEVCNTR0_EL0(2),
2841 	AMU_AMEVCNTR0_EL0(3),
2842 	AMU_AMEVCNTR0_EL0(4),
2843 	AMU_AMEVCNTR0_EL0(5),
2844 	AMU_AMEVCNTR0_EL0(6),
2845 	AMU_AMEVCNTR0_EL0(7),
2846 	AMU_AMEVCNTR0_EL0(8),
2847 	AMU_AMEVCNTR0_EL0(9),
2848 	AMU_AMEVCNTR0_EL0(10),
2849 	AMU_AMEVCNTR0_EL0(11),
2850 	AMU_AMEVCNTR0_EL0(12),
2851 	AMU_AMEVCNTR0_EL0(13),
2852 	AMU_AMEVCNTR0_EL0(14),
2853 	AMU_AMEVCNTR0_EL0(15),
2854 	AMU_AMEVTYPER0_EL0(0),
2855 	AMU_AMEVTYPER0_EL0(1),
2856 	AMU_AMEVTYPER0_EL0(2),
2857 	AMU_AMEVTYPER0_EL0(3),
2858 	AMU_AMEVTYPER0_EL0(4),
2859 	AMU_AMEVTYPER0_EL0(5),
2860 	AMU_AMEVTYPER0_EL0(6),
2861 	AMU_AMEVTYPER0_EL0(7),
2862 	AMU_AMEVTYPER0_EL0(8),
2863 	AMU_AMEVTYPER0_EL0(9),
2864 	AMU_AMEVTYPER0_EL0(10),
2865 	AMU_AMEVTYPER0_EL0(11),
2866 	AMU_AMEVTYPER0_EL0(12),
2867 	AMU_AMEVTYPER0_EL0(13),
2868 	AMU_AMEVTYPER0_EL0(14),
2869 	AMU_AMEVTYPER0_EL0(15),
2870 	AMU_AMEVCNTR1_EL0(0),
2871 	AMU_AMEVCNTR1_EL0(1),
2872 	AMU_AMEVCNTR1_EL0(2),
2873 	AMU_AMEVCNTR1_EL0(3),
2874 	AMU_AMEVCNTR1_EL0(4),
2875 	AMU_AMEVCNTR1_EL0(5),
2876 	AMU_AMEVCNTR1_EL0(6),
2877 	AMU_AMEVCNTR1_EL0(7),
2878 	AMU_AMEVCNTR1_EL0(8),
2879 	AMU_AMEVCNTR1_EL0(9),
2880 	AMU_AMEVCNTR1_EL0(10),
2881 	AMU_AMEVCNTR1_EL0(11),
2882 	AMU_AMEVCNTR1_EL0(12),
2883 	AMU_AMEVCNTR1_EL0(13),
2884 	AMU_AMEVCNTR1_EL0(14),
2885 	AMU_AMEVCNTR1_EL0(15),
2886 	AMU_AMEVTYPER1_EL0(0),
2887 	AMU_AMEVTYPER1_EL0(1),
2888 	AMU_AMEVTYPER1_EL0(2),
2889 	AMU_AMEVTYPER1_EL0(3),
2890 	AMU_AMEVTYPER1_EL0(4),
2891 	AMU_AMEVTYPER1_EL0(5),
2892 	AMU_AMEVTYPER1_EL0(6),
2893 	AMU_AMEVTYPER1_EL0(7),
2894 	AMU_AMEVTYPER1_EL0(8),
2895 	AMU_AMEVTYPER1_EL0(9),
2896 	AMU_AMEVTYPER1_EL0(10),
2897 	AMU_AMEVTYPER1_EL0(11),
2898 	AMU_AMEVTYPER1_EL0(12),
2899 	AMU_AMEVTYPER1_EL0(13),
2900 	AMU_AMEVTYPER1_EL0(14),
2901 	AMU_AMEVTYPER1_EL0(15),
2902 
2903 	{ SYS_DESC(SYS_CNTPCT_EL0), access_arch_timer },
2904 	{ SYS_DESC(SYS_CNTPCTSS_EL0), access_arch_timer },
2905 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
2906 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
2907 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
2908 
2909 	/* PMEVCNTRn_EL0 */
2910 	PMU_PMEVCNTR_EL0(0),
2911 	PMU_PMEVCNTR_EL0(1),
2912 	PMU_PMEVCNTR_EL0(2),
2913 	PMU_PMEVCNTR_EL0(3),
2914 	PMU_PMEVCNTR_EL0(4),
2915 	PMU_PMEVCNTR_EL0(5),
2916 	PMU_PMEVCNTR_EL0(6),
2917 	PMU_PMEVCNTR_EL0(7),
2918 	PMU_PMEVCNTR_EL0(8),
2919 	PMU_PMEVCNTR_EL0(9),
2920 	PMU_PMEVCNTR_EL0(10),
2921 	PMU_PMEVCNTR_EL0(11),
2922 	PMU_PMEVCNTR_EL0(12),
2923 	PMU_PMEVCNTR_EL0(13),
2924 	PMU_PMEVCNTR_EL0(14),
2925 	PMU_PMEVCNTR_EL0(15),
2926 	PMU_PMEVCNTR_EL0(16),
2927 	PMU_PMEVCNTR_EL0(17),
2928 	PMU_PMEVCNTR_EL0(18),
2929 	PMU_PMEVCNTR_EL0(19),
2930 	PMU_PMEVCNTR_EL0(20),
2931 	PMU_PMEVCNTR_EL0(21),
2932 	PMU_PMEVCNTR_EL0(22),
2933 	PMU_PMEVCNTR_EL0(23),
2934 	PMU_PMEVCNTR_EL0(24),
2935 	PMU_PMEVCNTR_EL0(25),
2936 	PMU_PMEVCNTR_EL0(26),
2937 	PMU_PMEVCNTR_EL0(27),
2938 	PMU_PMEVCNTR_EL0(28),
2939 	PMU_PMEVCNTR_EL0(29),
2940 	PMU_PMEVCNTR_EL0(30),
2941 	/* PMEVTYPERn_EL0 */
2942 	PMU_PMEVTYPER_EL0(0),
2943 	PMU_PMEVTYPER_EL0(1),
2944 	PMU_PMEVTYPER_EL0(2),
2945 	PMU_PMEVTYPER_EL0(3),
2946 	PMU_PMEVTYPER_EL0(4),
2947 	PMU_PMEVTYPER_EL0(5),
2948 	PMU_PMEVTYPER_EL0(6),
2949 	PMU_PMEVTYPER_EL0(7),
2950 	PMU_PMEVTYPER_EL0(8),
2951 	PMU_PMEVTYPER_EL0(9),
2952 	PMU_PMEVTYPER_EL0(10),
2953 	PMU_PMEVTYPER_EL0(11),
2954 	PMU_PMEVTYPER_EL0(12),
2955 	PMU_PMEVTYPER_EL0(13),
2956 	PMU_PMEVTYPER_EL0(14),
2957 	PMU_PMEVTYPER_EL0(15),
2958 	PMU_PMEVTYPER_EL0(16),
2959 	PMU_PMEVTYPER_EL0(17),
2960 	PMU_PMEVTYPER_EL0(18),
2961 	PMU_PMEVTYPER_EL0(19),
2962 	PMU_PMEVTYPER_EL0(20),
2963 	PMU_PMEVTYPER_EL0(21),
2964 	PMU_PMEVTYPER_EL0(22),
2965 	PMU_PMEVTYPER_EL0(23),
2966 	PMU_PMEVTYPER_EL0(24),
2967 	PMU_PMEVTYPER_EL0(25),
2968 	PMU_PMEVTYPER_EL0(26),
2969 	PMU_PMEVTYPER_EL0(27),
2970 	PMU_PMEVTYPER_EL0(28),
2971 	PMU_PMEVTYPER_EL0(29),
2972 	PMU_PMEVTYPER_EL0(30),
2973 	/*
2974 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
2975 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
2976 	 */
2977 	{ PMU_SYS_REG(PMCCFILTR_EL0), .access = access_pmu_evtyper,
2978 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
2979 
2980 	EL2_REG_VNCR(VPIDR_EL2, reset_unknown, 0),
2981 	EL2_REG_VNCR(VMPIDR_EL2, reset_unknown, 0),
2982 	EL2_REG(SCTLR_EL2, access_rw, reset_val, SCTLR_EL2_RES1),
2983 	EL2_REG(ACTLR_EL2, access_rw, reset_val, 0),
2984 	EL2_REG_VNCR(HCR_EL2, reset_hcr, 0),
2985 	EL2_REG(MDCR_EL2, access_rw, reset_val, 0),
2986 	EL2_REG(CPTR_EL2, access_rw, reset_val, CPTR_NVHE_EL2_RES1),
2987 	EL2_REG_VNCR(HSTR_EL2, reset_val, 0),
2988 	EL2_REG_VNCR(HFGRTR_EL2, reset_val, 0),
2989 	EL2_REG_VNCR(HFGWTR_EL2, reset_val, 0),
2990 	EL2_REG_VNCR(HFGITR_EL2, reset_val, 0),
2991 	EL2_REG_VNCR(HACR_EL2, reset_val, 0),
2992 
2993 	EL2_REG_FILTERED(ZCR_EL2, access_zcr_el2, reset_val, 0,
2994 			 sve_el2_visibility),
2995 
2996 	EL2_REG_VNCR(HCRX_EL2, reset_val, 0),
2997 
2998 	EL2_REG(TTBR0_EL2, access_rw, reset_val, 0),
2999 	EL2_REG(TTBR1_EL2, access_rw, reset_val, 0),
3000 	EL2_REG(TCR_EL2, access_rw, reset_val, TCR_EL2_RES1),
3001 	EL2_REG_FILTERED(TCR2_EL2, access_rw, reset_val, TCR2_EL2_RES1,
3002 			 tcr2_el2_visibility),
3003 	EL2_REG_VNCR(VTTBR_EL2, reset_val, 0),
3004 	EL2_REG_VNCR(VTCR_EL2, reset_val, 0),
3005 
3006 	{ SYS_DESC(SYS_DACR32_EL2), undef_access, reset_unknown, DACR32_EL2 },
3007 	EL2_REG_VNCR(HDFGRTR_EL2, reset_val, 0),
3008 	EL2_REG_VNCR(HDFGWTR_EL2, reset_val, 0),
3009 	EL2_REG_VNCR(HAFGRTR_EL2, reset_val, 0),
3010 	EL2_REG_REDIR(SPSR_EL2, reset_val, 0),
3011 	EL2_REG_REDIR(ELR_EL2, reset_val, 0),
3012 	{ SYS_DESC(SYS_SP_EL1), access_sp_el1},
3013 
3014 	/* AArch32 SPSR_* are RES0 if trapped from a NV guest */
3015 	{ SYS_DESC(SYS_SPSR_irq), .access = trap_raz_wi },
3016 	{ SYS_DESC(SYS_SPSR_abt), .access = trap_raz_wi },
3017 	{ SYS_DESC(SYS_SPSR_und), .access = trap_raz_wi },
3018 	{ SYS_DESC(SYS_SPSR_fiq), .access = trap_raz_wi },
3019 
3020 	{ SYS_DESC(SYS_IFSR32_EL2), undef_access, reset_unknown, IFSR32_EL2 },
3021 	EL2_REG(AFSR0_EL2, access_rw, reset_val, 0),
3022 	EL2_REG(AFSR1_EL2, access_rw, reset_val, 0),
3023 	EL2_REG_REDIR(ESR_EL2, reset_val, 0),
3024 	{ SYS_DESC(SYS_FPEXC32_EL2), undef_access, reset_val, FPEXC32_EL2, 0x700 },
3025 
3026 	EL2_REG_REDIR(FAR_EL2, reset_val, 0),
3027 	EL2_REG(HPFAR_EL2, access_rw, reset_val, 0),
3028 
3029 	EL2_REG(MAIR_EL2, access_rw, reset_val, 0),
3030 	EL2_REG_FILTERED(PIRE0_EL2, access_rw, reset_val, 0,
3031 			 s1pie_el2_visibility),
3032 	EL2_REG_FILTERED(PIR_EL2, access_rw, reset_val, 0,
3033 			 s1pie_el2_visibility),
3034 	EL2_REG_FILTERED(POR_EL2, access_rw, reset_val, 0,
3035 			 s1poe_el2_visibility),
3036 	EL2_REG(AMAIR_EL2, access_rw, reset_val, 0),
3037 	{ SYS_DESC(SYS_MPAMHCR_EL2), undef_access },
3038 	{ SYS_DESC(SYS_MPAMVPMV_EL2), undef_access },
3039 	{ SYS_DESC(SYS_MPAM2_EL2), undef_access },
3040 	{ SYS_DESC(SYS_MPAMVPM0_EL2), undef_access },
3041 	{ SYS_DESC(SYS_MPAMVPM1_EL2), undef_access },
3042 	{ SYS_DESC(SYS_MPAMVPM2_EL2), undef_access },
3043 	{ SYS_DESC(SYS_MPAMVPM3_EL2), undef_access },
3044 	{ SYS_DESC(SYS_MPAMVPM4_EL2), undef_access },
3045 	{ SYS_DESC(SYS_MPAMVPM5_EL2), undef_access },
3046 	{ SYS_DESC(SYS_MPAMVPM6_EL2), undef_access },
3047 	{ SYS_DESC(SYS_MPAMVPM7_EL2), undef_access },
3048 
3049 	EL2_REG(VBAR_EL2, access_rw, reset_val, 0),
3050 	EL2_REG(RVBAR_EL2, access_rw, reset_val, 0),
3051 	{ SYS_DESC(SYS_RMR_EL2), undef_access },
3052 
3053 	EL2_REG_VNCR(ICH_HCR_EL2, reset_val, 0),
3054 
3055 	EL2_REG(CONTEXTIDR_EL2, access_rw, reset_val, 0),
3056 	EL2_REG(TPIDR_EL2, access_rw, reset_val, 0),
3057 
3058 	EL2_REG_VNCR(CNTVOFF_EL2, reset_val, 0),
3059 	EL2_REG(CNTHCTL_EL2, access_rw, reset_val, 0),
3060 
3061 	{ SYS_DESC(SYS_CNTKCTL_EL12), access_cntkctl_el12 },
3062 
3063 	EL2_REG(SP_EL2, NULL, reset_unknown, 0),
3064 };
3065 
3066 static bool handle_at_s1e01(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3067 			    const struct sys_reg_desc *r)
3068 {
3069 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3070 
3071 	__kvm_at_s1e01(vcpu, op, p->regval);
3072 
3073 	return true;
3074 }
3075 
3076 static bool handle_at_s1e2(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3077 			   const struct sys_reg_desc *r)
3078 {
3079 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3080 
3081 	/* There is no FGT associated with AT S1E2A :-( */
3082 	if (op == OP_AT_S1E2A &&
3083 	    !kvm_has_feat(vcpu->kvm, ID_AA64ISAR2_EL1, ATS1A, IMP)) {
3084 		kvm_inject_undefined(vcpu);
3085 		return false;
3086 	}
3087 
3088 	__kvm_at_s1e2(vcpu, op, p->regval);
3089 
3090 	return true;
3091 }
3092 
3093 static bool handle_at_s12(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3094 			  const struct sys_reg_desc *r)
3095 {
3096 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3097 
3098 	__kvm_at_s12(vcpu, op, p->regval);
3099 
3100 	return true;
3101 }
3102 
3103 static bool kvm_supported_tlbi_s12_op(struct kvm_vcpu *vpcu, u32 instr)
3104 {
3105 	struct kvm *kvm = vpcu->kvm;
3106 	u8 CRm = sys_reg_CRm(instr);
3107 
3108 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
3109 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
3110 		return false;
3111 
3112 	if (CRm == TLBI_CRm_nROS &&
3113 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
3114 		return false;
3115 
3116 	return true;
3117 }
3118 
3119 static bool handle_alle1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3120 			   const struct sys_reg_desc *r)
3121 {
3122 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3123 
3124 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
3125 		return undef_access(vcpu, p, r);
3126 
3127 	write_lock(&vcpu->kvm->mmu_lock);
3128 
3129 	/*
3130 	 * Drop all shadow S2s, resulting in S1/S2 TLBIs for each of the
3131 	 * corresponding VMIDs.
3132 	 */
3133 	kvm_nested_s2_unmap(vcpu->kvm, true);
3134 
3135 	write_unlock(&vcpu->kvm->mmu_lock);
3136 
3137 	return true;
3138 }
3139 
3140 static bool kvm_supported_tlbi_ipas2_op(struct kvm_vcpu *vpcu, u32 instr)
3141 {
3142 	struct kvm *kvm = vpcu->kvm;
3143 	u8 CRm = sys_reg_CRm(instr);
3144 	u8 Op2 = sys_reg_Op2(instr);
3145 
3146 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
3147 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
3148 		return false;
3149 
3150 	if (CRm == TLBI_CRm_IPAIS && (Op2 == 2 || Op2 == 6) &&
3151 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3152 		return false;
3153 
3154 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 0 || Op2 == 4) &&
3155 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
3156 		return false;
3157 
3158 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 3 || Op2 == 7) &&
3159 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3160 		return false;
3161 
3162 	return true;
3163 }
3164 
3165 /* Only defined here as this is an internal "abstraction" */
3166 union tlbi_info {
3167 	struct {
3168 		u64	start;
3169 		u64	size;
3170 	} range;
3171 
3172 	struct {
3173 		u64	addr;
3174 	} ipa;
3175 
3176 	struct {
3177 		u64	addr;
3178 		u32	encoding;
3179 	} va;
3180 };
3181 
3182 static void s2_mmu_unmap_range(struct kvm_s2_mmu *mmu,
3183 			       const union tlbi_info *info)
3184 {
3185 	/*
3186 	 * The unmap operation is allowed to drop the MMU lock and block, which
3187 	 * means that @mmu could be used for a different context than the one
3188 	 * currently being invalidated.
3189 	 *
3190 	 * This behavior is still safe, as:
3191 	 *
3192 	 *  1) The vCPU(s) that recycled the MMU are responsible for invalidating
3193 	 *     the entire MMU before reusing it, which still honors the intent
3194 	 *     of a TLBI.
3195 	 *
3196 	 *  2) Until the guest TLBI instruction is 'retired' (i.e. increment PC
3197 	 *     and ERET to the guest), other vCPUs are allowed to use stale
3198 	 *     translations.
3199 	 *
3200 	 *  3) Accidentally unmapping an unrelated MMU context is nonfatal, and
3201 	 *     at worst may cause more aborts for shadow stage-2 fills.
3202 	 *
3203 	 * Dropping the MMU lock also implies that shadow stage-2 fills could
3204 	 * happen behind the back of the TLBI. This is still safe, though, as
3205 	 * the L1 needs to put its stage-2 in a consistent state before doing
3206 	 * the TLBI.
3207 	 */
3208 	kvm_stage2_unmap_range(mmu, info->range.start, info->range.size, true);
3209 }
3210 
3211 static bool handle_vmalls12e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3212 				const struct sys_reg_desc *r)
3213 {
3214 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3215 	u64 limit, vttbr;
3216 
3217 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
3218 		return undef_access(vcpu, p, r);
3219 
3220 	vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3221 	limit = BIT_ULL(kvm_get_pa_bits(vcpu->kvm));
3222 
3223 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3224 				   &(union tlbi_info) {
3225 					   .range = {
3226 						   .start = 0,
3227 						   .size = limit,
3228 					   },
3229 				   },
3230 				   s2_mmu_unmap_range);
3231 
3232 	return true;
3233 }
3234 
3235 static bool handle_ripas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3236 			      const struct sys_reg_desc *r)
3237 {
3238 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3239 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3240 	u64 base, range, tg, num, scale;
3241 	int shift;
3242 
3243 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
3244 		return undef_access(vcpu, p, r);
3245 
3246 	/*
3247 	 * Because the shadow S2 structure doesn't necessarily reflect that
3248 	 * of the guest's S2 (different base granule size, for example), we
3249 	 * decide to ignore TTL and only use the described range.
3250 	 */
3251 	tg	= FIELD_GET(GENMASK(47, 46), p->regval);
3252 	scale	= FIELD_GET(GENMASK(45, 44), p->regval);
3253 	num	= FIELD_GET(GENMASK(43, 39), p->regval);
3254 	base	= p->regval & GENMASK(36, 0);
3255 
3256 	switch(tg) {
3257 	case 1:
3258 		shift = 12;
3259 		break;
3260 	case 2:
3261 		shift = 14;
3262 		break;
3263 	case 3:
3264 	default:		/* IMPDEF: handle tg==0 as 64k */
3265 		shift = 16;
3266 		break;
3267 	}
3268 
3269 	base <<= shift;
3270 	range = __TLBI_RANGE_PAGES(num, scale) << shift;
3271 
3272 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3273 				   &(union tlbi_info) {
3274 					   .range = {
3275 						   .start = base,
3276 						   .size = range,
3277 					   },
3278 				   },
3279 				   s2_mmu_unmap_range);
3280 
3281 	return true;
3282 }
3283 
3284 static void s2_mmu_unmap_ipa(struct kvm_s2_mmu *mmu,
3285 			     const union tlbi_info *info)
3286 {
3287 	unsigned long max_size;
3288 	u64 base_addr;
3289 
3290 	/*
3291 	 * We drop a number of things from the supplied value:
3292 	 *
3293 	 * - NS bit: we're non-secure only.
3294 	 *
3295 	 * - IPA[51:48]: We don't support 52bit IPA just yet...
3296 	 *
3297 	 * And of course, adjust the IPA to be on an actual address.
3298 	 */
3299 	base_addr = (info->ipa.addr & GENMASK_ULL(35, 0)) << 12;
3300 	max_size = compute_tlb_inval_range(mmu, info->ipa.addr);
3301 	base_addr &= ~(max_size - 1);
3302 
3303 	/*
3304 	 * See comment in s2_mmu_unmap_range() for why this is allowed to
3305 	 * reschedule.
3306 	 */
3307 	kvm_stage2_unmap_range(mmu, base_addr, max_size, true);
3308 }
3309 
3310 static bool handle_ipas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3311 			     const struct sys_reg_desc *r)
3312 {
3313 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3314 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3315 
3316 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
3317 		return undef_access(vcpu, p, r);
3318 
3319 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3320 				   &(union tlbi_info) {
3321 					   .ipa = {
3322 						   .addr = p->regval,
3323 					   },
3324 				   },
3325 				   s2_mmu_unmap_ipa);
3326 
3327 	return true;
3328 }
3329 
3330 static void s2_mmu_tlbi_s1e1(struct kvm_s2_mmu *mmu,
3331 			     const union tlbi_info *info)
3332 {
3333 	WARN_ON(__kvm_tlbi_s1e2(mmu, info->va.addr, info->va.encoding));
3334 }
3335 
3336 static bool handle_tlbi_el1(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3337 			    const struct sys_reg_desc *r)
3338 {
3339 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3340 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3341 
3342 	/*
3343 	 * If we're here, this is because we've trapped on a EL1 TLBI
3344 	 * instruction that affects the EL1 translation regime while
3345 	 * we're running in a context that doesn't allow us to let the
3346 	 * HW do its thing (aka vEL2):
3347 	 *
3348 	 * - HCR_EL2.E2H == 0 : a non-VHE guest
3349 	 * - HCR_EL2.{E2H,TGE} == { 1, 0 } : a VHE guest in guest mode
3350 	 *
3351 	 * We don't expect these helpers to ever be called when running
3352 	 * in a vEL1 context.
3353 	 */
3354 
3355 	WARN_ON(!vcpu_is_el2(vcpu));
3356 
3357 	if (!kvm_supported_tlbi_s1e1_op(vcpu, sys_encoding))
3358 		return undef_access(vcpu, p, r);
3359 
3360 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3361 				   &(union tlbi_info) {
3362 					   .va = {
3363 						   .addr = p->regval,
3364 						   .encoding = sys_encoding,
3365 					   },
3366 				   },
3367 				   s2_mmu_tlbi_s1e1);
3368 
3369 	return true;
3370 }
3371 
3372 #define SYS_INSN(insn, access_fn)					\
3373 	{								\
3374 		SYS_DESC(OP_##insn),					\
3375 		.access = (access_fn),					\
3376 	}
3377 
3378 static struct sys_reg_desc sys_insn_descs[] = {
3379 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
3380 	{ SYS_DESC(SYS_DC_IGSW), access_dcgsw },
3381 	{ SYS_DESC(SYS_DC_IGDSW), access_dcgsw },
3382 
3383 	SYS_INSN(AT_S1E1R, handle_at_s1e01),
3384 	SYS_INSN(AT_S1E1W, handle_at_s1e01),
3385 	SYS_INSN(AT_S1E0R, handle_at_s1e01),
3386 	SYS_INSN(AT_S1E0W, handle_at_s1e01),
3387 	SYS_INSN(AT_S1E1RP, handle_at_s1e01),
3388 	SYS_INSN(AT_S1E1WP, handle_at_s1e01),
3389 
3390 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
3391 	{ SYS_DESC(SYS_DC_CGSW), access_dcgsw },
3392 	{ SYS_DESC(SYS_DC_CGDSW), access_dcgsw },
3393 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
3394 	{ SYS_DESC(SYS_DC_CIGSW), access_dcgsw },
3395 	{ SYS_DESC(SYS_DC_CIGDSW), access_dcgsw },
3396 
3397 	SYS_INSN(TLBI_VMALLE1OS, handle_tlbi_el1),
3398 	SYS_INSN(TLBI_VAE1OS, handle_tlbi_el1),
3399 	SYS_INSN(TLBI_ASIDE1OS, handle_tlbi_el1),
3400 	SYS_INSN(TLBI_VAAE1OS, handle_tlbi_el1),
3401 	SYS_INSN(TLBI_VALE1OS, handle_tlbi_el1),
3402 	SYS_INSN(TLBI_VAALE1OS, handle_tlbi_el1),
3403 
3404 	SYS_INSN(TLBI_RVAE1IS, handle_tlbi_el1),
3405 	SYS_INSN(TLBI_RVAAE1IS, handle_tlbi_el1),
3406 	SYS_INSN(TLBI_RVALE1IS, handle_tlbi_el1),
3407 	SYS_INSN(TLBI_RVAALE1IS, handle_tlbi_el1),
3408 
3409 	SYS_INSN(TLBI_VMALLE1IS, handle_tlbi_el1),
3410 	SYS_INSN(TLBI_VAE1IS, handle_tlbi_el1),
3411 	SYS_INSN(TLBI_ASIDE1IS, handle_tlbi_el1),
3412 	SYS_INSN(TLBI_VAAE1IS, handle_tlbi_el1),
3413 	SYS_INSN(TLBI_VALE1IS, handle_tlbi_el1),
3414 	SYS_INSN(TLBI_VAALE1IS, handle_tlbi_el1),
3415 
3416 	SYS_INSN(TLBI_RVAE1OS, handle_tlbi_el1),
3417 	SYS_INSN(TLBI_RVAAE1OS, handle_tlbi_el1),
3418 	SYS_INSN(TLBI_RVALE1OS, handle_tlbi_el1),
3419 	SYS_INSN(TLBI_RVAALE1OS, handle_tlbi_el1),
3420 
3421 	SYS_INSN(TLBI_RVAE1, handle_tlbi_el1),
3422 	SYS_INSN(TLBI_RVAAE1, handle_tlbi_el1),
3423 	SYS_INSN(TLBI_RVALE1, handle_tlbi_el1),
3424 	SYS_INSN(TLBI_RVAALE1, handle_tlbi_el1),
3425 
3426 	SYS_INSN(TLBI_VMALLE1, handle_tlbi_el1),
3427 	SYS_INSN(TLBI_VAE1, handle_tlbi_el1),
3428 	SYS_INSN(TLBI_ASIDE1, handle_tlbi_el1),
3429 	SYS_INSN(TLBI_VAAE1, handle_tlbi_el1),
3430 	SYS_INSN(TLBI_VALE1, handle_tlbi_el1),
3431 	SYS_INSN(TLBI_VAALE1, handle_tlbi_el1),
3432 
3433 	SYS_INSN(TLBI_VMALLE1OSNXS, handle_tlbi_el1),
3434 	SYS_INSN(TLBI_VAE1OSNXS, handle_tlbi_el1),
3435 	SYS_INSN(TLBI_ASIDE1OSNXS, handle_tlbi_el1),
3436 	SYS_INSN(TLBI_VAAE1OSNXS, handle_tlbi_el1),
3437 	SYS_INSN(TLBI_VALE1OSNXS, handle_tlbi_el1),
3438 	SYS_INSN(TLBI_VAALE1OSNXS, handle_tlbi_el1),
3439 
3440 	SYS_INSN(TLBI_RVAE1ISNXS, handle_tlbi_el1),
3441 	SYS_INSN(TLBI_RVAAE1ISNXS, handle_tlbi_el1),
3442 	SYS_INSN(TLBI_RVALE1ISNXS, handle_tlbi_el1),
3443 	SYS_INSN(TLBI_RVAALE1ISNXS, handle_tlbi_el1),
3444 
3445 	SYS_INSN(TLBI_VMALLE1ISNXS, handle_tlbi_el1),
3446 	SYS_INSN(TLBI_VAE1ISNXS, handle_tlbi_el1),
3447 	SYS_INSN(TLBI_ASIDE1ISNXS, handle_tlbi_el1),
3448 	SYS_INSN(TLBI_VAAE1ISNXS, handle_tlbi_el1),
3449 	SYS_INSN(TLBI_VALE1ISNXS, handle_tlbi_el1),
3450 	SYS_INSN(TLBI_VAALE1ISNXS, handle_tlbi_el1),
3451 
3452 	SYS_INSN(TLBI_RVAE1OSNXS, handle_tlbi_el1),
3453 	SYS_INSN(TLBI_RVAAE1OSNXS, handle_tlbi_el1),
3454 	SYS_INSN(TLBI_RVALE1OSNXS, handle_tlbi_el1),
3455 	SYS_INSN(TLBI_RVAALE1OSNXS, handle_tlbi_el1),
3456 
3457 	SYS_INSN(TLBI_RVAE1NXS, handle_tlbi_el1),
3458 	SYS_INSN(TLBI_RVAAE1NXS, handle_tlbi_el1),
3459 	SYS_INSN(TLBI_RVALE1NXS, handle_tlbi_el1),
3460 	SYS_INSN(TLBI_RVAALE1NXS, handle_tlbi_el1),
3461 
3462 	SYS_INSN(TLBI_VMALLE1NXS, handle_tlbi_el1),
3463 	SYS_INSN(TLBI_VAE1NXS, handle_tlbi_el1),
3464 	SYS_INSN(TLBI_ASIDE1NXS, handle_tlbi_el1),
3465 	SYS_INSN(TLBI_VAAE1NXS, handle_tlbi_el1),
3466 	SYS_INSN(TLBI_VALE1NXS, handle_tlbi_el1),
3467 	SYS_INSN(TLBI_VAALE1NXS, handle_tlbi_el1),
3468 
3469 	SYS_INSN(AT_S1E2R, handle_at_s1e2),
3470 	SYS_INSN(AT_S1E2W, handle_at_s1e2),
3471 	SYS_INSN(AT_S12E1R, handle_at_s12),
3472 	SYS_INSN(AT_S12E1W, handle_at_s12),
3473 	SYS_INSN(AT_S12E0R, handle_at_s12),
3474 	SYS_INSN(AT_S12E0W, handle_at_s12),
3475 	SYS_INSN(AT_S1E2A, handle_at_s1e2),
3476 
3477 	SYS_INSN(TLBI_IPAS2E1IS, handle_ipas2e1is),
3478 	SYS_INSN(TLBI_RIPAS2E1IS, handle_ripas2e1is),
3479 	SYS_INSN(TLBI_IPAS2LE1IS, handle_ipas2e1is),
3480 	SYS_INSN(TLBI_RIPAS2LE1IS, handle_ripas2e1is),
3481 
3482 	SYS_INSN(TLBI_ALLE2OS, undef_access),
3483 	SYS_INSN(TLBI_VAE2OS, undef_access),
3484 	SYS_INSN(TLBI_ALLE1OS, handle_alle1is),
3485 	SYS_INSN(TLBI_VALE2OS, undef_access),
3486 	SYS_INSN(TLBI_VMALLS12E1OS, handle_vmalls12e1is),
3487 
3488 	SYS_INSN(TLBI_RVAE2IS, undef_access),
3489 	SYS_INSN(TLBI_RVALE2IS, undef_access),
3490 
3491 	SYS_INSN(TLBI_ALLE1IS, handle_alle1is),
3492 	SYS_INSN(TLBI_VMALLS12E1IS, handle_vmalls12e1is),
3493 	SYS_INSN(TLBI_IPAS2E1OS, handle_ipas2e1is),
3494 	SYS_INSN(TLBI_IPAS2E1, handle_ipas2e1is),
3495 	SYS_INSN(TLBI_RIPAS2E1, handle_ripas2e1is),
3496 	SYS_INSN(TLBI_RIPAS2E1OS, handle_ripas2e1is),
3497 	SYS_INSN(TLBI_IPAS2LE1OS, handle_ipas2e1is),
3498 	SYS_INSN(TLBI_IPAS2LE1, handle_ipas2e1is),
3499 	SYS_INSN(TLBI_RIPAS2LE1, handle_ripas2e1is),
3500 	SYS_INSN(TLBI_RIPAS2LE1OS, handle_ripas2e1is),
3501 	SYS_INSN(TLBI_RVAE2OS, undef_access),
3502 	SYS_INSN(TLBI_RVALE2OS, undef_access),
3503 	SYS_INSN(TLBI_RVAE2, undef_access),
3504 	SYS_INSN(TLBI_RVALE2, undef_access),
3505 	SYS_INSN(TLBI_ALLE1, handle_alle1is),
3506 	SYS_INSN(TLBI_VMALLS12E1, handle_vmalls12e1is),
3507 
3508 	SYS_INSN(TLBI_IPAS2E1ISNXS, handle_ipas2e1is),
3509 	SYS_INSN(TLBI_RIPAS2E1ISNXS, handle_ripas2e1is),
3510 	SYS_INSN(TLBI_IPAS2LE1ISNXS, handle_ipas2e1is),
3511 	SYS_INSN(TLBI_RIPAS2LE1ISNXS, handle_ripas2e1is),
3512 
3513 	SYS_INSN(TLBI_ALLE2OSNXS, undef_access),
3514 	SYS_INSN(TLBI_VAE2OSNXS, undef_access),
3515 	SYS_INSN(TLBI_ALLE1OSNXS, handle_alle1is),
3516 	SYS_INSN(TLBI_VALE2OSNXS, undef_access),
3517 	SYS_INSN(TLBI_VMALLS12E1OSNXS, handle_vmalls12e1is),
3518 
3519 	SYS_INSN(TLBI_RVAE2ISNXS, undef_access),
3520 	SYS_INSN(TLBI_RVALE2ISNXS, undef_access),
3521 	SYS_INSN(TLBI_ALLE2ISNXS, undef_access),
3522 	SYS_INSN(TLBI_VAE2ISNXS, undef_access),
3523 
3524 	SYS_INSN(TLBI_ALLE1ISNXS, handle_alle1is),
3525 	SYS_INSN(TLBI_VALE2ISNXS, undef_access),
3526 	SYS_INSN(TLBI_VMALLS12E1ISNXS, handle_vmalls12e1is),
3527 	SYS_INSN(TLBI_IPAS2E1OSNXS, handle_ipas2e1is),
3528 	SYS_INSN(TLBI_IPAS2E1NXS, handle_ipas2e1is),
3529 	SYS_INSN(TLBI_RIPAS2E1NXS, handle_ripas2e1is),
3530 	SYS_INSN(TLBI_RIPAS2E1OSNXS, handle_ripas2e1is),
3531 	SYS_INSN(TLBI_IPAS2LE1OSNXS, handle_ipas2e1is),
3532 	SYS_INSN(TLBI_IPAS2LE1NXS, handle_ipas2e1is),
3533 	SYS_INSN(TLBI_RIPAS2LE1NXS, handle_ripas2e1is),
3534 	SYS_INSN(TLBI_RIPAS2LE1OSNXS, handle_ripas2e1is),
3535 	SYS_INSN(TLBI_RVAE2OSNXS, undef_access),
3536 	SYS_INSN(TLBI_RVALE2OSNXS, undef_access),
3537 	SYS_INSN(TLBI_RVAE2NXS, undef_access),
3538 	SYS_INSN(TLBI_RVALE2NXS, undef_access),
3539 	SYS_INSN(TLBI_ALLE2NXS, undef_access),
3540 	SYS_INSN(TLBI_VAE2NXS, undef_access),
3541 	SYS_INSN(TLBI_ALLE1NXS, handle_alle1is),
3542 	SYS_INSN(TLBI_VALE2NXS, undef_access),
3543 	SYS_INSN(TLBI_VMALLS12E1NXS, handle_vmalls12e1is),
3544 };
3545 
3546 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
3547 			struct sys_reg_params *p,
3548 			const struct sys_reg_desc *r)
3549 {
3550 	if (p->is_write) {
3551 		return ignore_write(vcpu, p);
3552 	} else {
3553 		u64 dfr = kvm_read_vm_id_reg(vcpu->kvm, SYS_ID_AA64DFR0_EL1);
3554 		u32 el3 = kvm_has_feat(vcpu->kvm, ID_AA64PFR0_EL1, EL3, IMP);
3555 
3556 		p->regval = ((SYS_FIELD_GET(ID_AA64DFR0_EL1, WRPs, dfr) << 28) |
3557 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, BRPs, dfr) << 24) |
3558 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, CTX_CMPs, dfr) << 20) |
3559 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, dfr) << 16) |
3560 			     (1 << 15) | (el3 << 14) | (el3 << 12));
3561 		return true;
3562 	}
3563 }
3564 
3565 /*
3566  * AArch32 debug register mappings
3567  *
3568  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
3569  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
3570  *
3571  * None of the other registers share their location, so treat them as
3572  * if they were 64bit.
3573  */
3574 #define DBG_BCR_BVR_WCR_WVR(n)						      \
3575 	/* DBGBVRn */							      \
3576 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, \
3577 	/* DBGBCRn */							      \
3578 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	      \
3579 	/* DBGWVRn */							      \
3580 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	      \
3581 	/* DBGWCRn */							      \
3582 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
3583 
3584 #define DBGBXVR(n)							      \
3585 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_bvr, NULL, n }
3586 
3587 /*
3588  * Trapped cp14 registers. We generally ignore most of the external
3589  * debug, on the principle that they don't really make sense to a
3590  * guest. Revisit this one day, would this principle change.
3591  */
3592 static const struct sys_reg_desc cp14_regs[] = {
3593 	/* DBGDIDR */
3594 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
3595 	/* DBGDTRRXext */
3596 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
3597 
3598 	DBG_BCR_BVR_WCR_WVR(0),
3599 	/* DBGDSCRint */
3600 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
3601 	DBG_BCR_BVR_WCR_WVR(1),
3602 	/* DBGDCCINT */
3603 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
3604 	/* DBGDSCRext */
3605 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
3606 	DBG_BCR_BVR_WCR_WVR(2),
3607 	/* DBGDTR[RT]Xint */
3608 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
3609 	/* DBGDTR[RT]Xext */
3610 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
3611 	DBG_BCR_BVR_WCR_WVR(3),
3612 	DBG_BCR_BVR_WCR_WVR(4),
3613 	DBG_BCR_BVR_WCR_WVR(5),
3614 	/* DBGWFAR */
3615 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
3616 	/* DBGOSECCR */
3617 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
3618 	DBG_BCR_BVR_WCR_WVR(6),
3619 	/* DBGVCR */
3620 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
3621 	DBG_BCR_BVR_WCR_WVR(7),
3622 	DBG_BCR_BVR_WCR_WVR(8),
3623 	DBG_BCR_BVR_WCR_WVR(9),
3624 	DBG_BCR_BVR_WCR_WVR(10),
3625 	DBG_BCR_BVR_WCR_WVR(11),
3626 	DBG_BCR_BVR_WCR_WVR(12),
3627 	DBG_BCR_BVR_WCR_WVR(13),
3628 	DBG_BCR_BVR_WCR_WVR(14),
3629 	DBG_BCR_BVR_WCR_WVR(15),
3630 
3631 	/* DBGDRAR (32bit) */
3632 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
3633 
3634 	DBGBXVR(0),
3635 	/* DBGOSLAR */
3636 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_oslar_el1 },
3637 	DBGBXVR(1),
3638 	/* DBGOSLSR */
3639 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1, NULL, OSLSR_EL1 },
3640 	DBGBXVR(2),
3641 	DBGBXVR(3),
3642 	/* DBGOSDLR */
3643 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
3644 	DBGBXVR(4),
3645 	/* DBGPRCR */
3646 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
3647 	DBGBXVR(5),
3648 	DBGBXVR(6),
3649 	DBGBXVR(7),
3650 	DBGBXVR(8),
3651 	DBGBXVR(9),
3652 	DBGBXVR(10),
3653 	DBGBXVR(11),
3654 	DBGBXVR(12),
3655 	DBGBXVR(13),
3656 	DBGBXVR(14),
3657 	DBGBXVR(15),
3658 
3659 	/* DBGDSAR (32bit) */
3660 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
3661 
3662 	/* DBGDEVID2 */
3663 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
3664 	/* DBGDEVID1 */
3665 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
3666 	/* DBGDEVID */
3667 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
3668 	/* DBGCLAIMSET */
3669 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
3670 	/* DBGCLAIMCLR */
3671 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
3672 	/* DBGAUTHSTATUS */
3673 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
3674 };
3675 
3676 /* Trapped cp14 64bit registers */
3677 static const struct sys_reg_desc cp14_64_regs[] = {
3678 	/* DBGDRAR (64bit) */
3679 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
3680 
3681 	/* DBGDSAR (64bit) */
3682 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
3683 };
3684 
3685 #define CP15_PMU_SYS_REG(_map, _Op1, _CRn, _CRm, _Op2)			\
3686 	AA32(_map),							\
3687 	Op1(_Op1), CRn(_CRn), CRm(_CRm), Op2(_Op2),			\
3688 	.visibility = pmu_visibility
3689 
3690 /* Macro to expand the PMEVCNTRn register */
3691 #define PMU_PMEVCNTR(n)							\
3692 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
3693 	  (0b1000 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
3694 	  .access = access_pmu_evcntr }
3695 
3696 /* Macro to expand the PMEVTYPERn register */
3697 #define PMU_PMEVTYPER(n)						\
3698 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
3699 	  (0b1100 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
3700 	  .access = access_pmu_evtyper }
3701 /*
3702  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
3703  * depending on the way they are accessed (as a 32bit or a 64bit
3704  * register).
3705  */
3706 static const struct sys_reg_desc cp15_regs[] = {
3707 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
3708 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
3709 	/* ACTLR */
3710 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
3711 	/* ACTLR2 */
3712 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
3713 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
3714 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
3715 	/* TTBCR */
3716 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
3717 	/* TTBCR2 */
3718 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
3719 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
3720 	{ CP15_SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
3721 	/* DFSR */
3722 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
3723 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
3724 	/* ADFSR */
3725 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
3726 	/* AIFSR */
3727 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
3728 	/* DFAR */
3729 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
3730 	/* IFAR */
3731 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
3732 
3733 	/*
3734 	 * DC{C,I,CI}SW operations:
3735 	 */
3736 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
3737 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
3738 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
3739 
3740 	/* PMU */
3741 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 0), .access = access_pmcr },
3742 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 1), .access = access_pmcnten },
3743 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 2), .access = access_pmcnten },
3744 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 3), .access = access_pmovs },
3745 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 4), .access = access_pmswinc },
3746 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 5), .access = access_pmselr },
3747 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 6), .access = access_pmceid },
3748 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 7), .access = access_pmceid },
3749 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 0), .access = access_pmu_evcntr },
3750 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 1), .access = access_pmu_evtyper },
3751 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 2), .access = access_pmu_evcntr },
3752 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 0), .access = access_pmuserenr },
3753 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 1), .access = access_pminten },
3754 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 2), .access = access_pminten },
3755 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 3), .access = access_pmovs },
3756 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 4), .access = access_pmceid },
3757 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 5), .access = access_pmceid },
3758 	/* PMMIR */
3759 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 6), .access = trap_raz_wi },
3760 
3761 	/* PRRR/MAIR0 */
3762 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
3763 	/* NMRR/MAIR1 */
3764 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
3765 	/* AMAIR0 */
3766 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
3767 	/* AMAIR1 */
3768 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
3769 
3770 	{ CP15_SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
3771 	{ CP15_SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
3772 	{ CP15_SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
3773 	{ CP15_SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
3774 	{ CP15_SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
3775 	{ CP15_SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
3776 	{ CP15_SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
3777 	{ CP15_SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
3778 	{ CP15_SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
3779 	{ CP15_SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
3780 	{ CP15_SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
3781 	{ CP15_SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
3782 	{ CP15_SYS_DESC(SYS_ICC_DIR_EL1), undef_access },
3783 	{ CP15_SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
3784 	{ CP15_SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
3785 	{ CP15_SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
3786 	{ CP15_SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
3787 	{ CP15_SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
3788 	{ CP15_SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
3789 	{ CP15_SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
3790 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
3791 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
3792 
3793 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
3794 
3795 	/* Arch Tmers */
3796 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
3797 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
3798 
3799 	/* PMEVCNTRn */
3800 	PMU_PMEVCNTR(0),
3801 	PMU_PMEVCNTR(1),
3802 	PMU_PMEVCNTR(2),
3803 	PMU_PMEVCNTR(3),
3804 	PMU_PMEVCNTR(4),
3805 	PMU_PMEVCNTR(5),
3806 	PMU_PMEVCNTR(6),
3807 	PMU_PMEVCNTR(7),
3808 	PMU_PMEVCNTR(8),
3809 	PMU_PMEVCNTR(9),
3810 	PMU_PMEVCNTR(10),
3811 	PMU_PMEVCNTR(11),
3812 	PMU_PMEVCNTR(12),
3813 	PMU_PMEVCNTR(13),
3814 	PMU_PMEVCNTR(14),
3815 	PMU_PMEVCNTR(15),
3816 	PMU_PMEVCNTR(16),
3817 	PMU_PMEVCNTR(17),
3818 	PMU_PMEVCNTR(18),
3819 	PMU_PMEVCNTR(19),
3820 	PMU_PMEVCNTR(20),
3821 	PMU_PMEVCNTR(21),
3822 	PMU_PMEVCNTR(22),
3823 	PMU_PMEVCNTR(23),
3824 	PMU_PMEVCNTR(24),
3825 	PMU_PMEVCNTR(25),
3826 	PMU_PMEVCNTR(26),
3827 	PMU_PMEVCNTR(27),
3828 	PMU_PMEVCNTR(28),
3829 	PMU_PMEVCNTR(29),
3830 	PMU_PMEVCNTR(30),
3831 	/* PMEVTYPERn */
3832 	PMU_PMEVTYPER(0),
3833 	PMU_PMEVTYPER(1),
3834 	PMU_PMEVTYPER(2),
3835 	PMU_PMEVTYPER(3),
3836 	PMU_PMEVTYPER(4),
3837 	PMU_PMEVTYPER(5),
3838 	PMU_PMEVTYPER(6),
3839 	PMU_PMEVTYPER(7),
3840 	PMU_PMEVTYPER(8),
3841 	PMU_PMEVTYPER(9),
3842 	PMU_PMEVTYPER(10),
3843 	PMU_PMEVTYPER(11),
3844 	PMU_PMEVTYPER(12),
3845 	PMU_PMEVTYPER(13),
3846 	PMU_PMEVTYPER(14),
3847 	PMU_PMEVTYPER(15),
3848 	PMU_PMEVTYPER(16),
3849 	PMU_PMEVTYPER(17),
3850 	PMU_PMEVTYPER(18),
3851 	PMU_PMEVTYPER(19),
3852 	PMU_PMEVTYPER(20),
3853 	PMU_PMEVTYPER(21),
3854 	PMU_PMEVTYPER(22),
3855 	PMU_PMEVTYPER(23),
3856 	PMU_PMEVTYPER(24),
3857 	PMU_PMEVTYPER(25),
3858 	PMU_PMEVTYPER(26),
3859 	PMU_PMEVTYPER(27),
3860 	PMU_PMEVTYPER(28),
3861 	PMU_PMEVTYPER(29),
3862 	PMU_PMEVTYPER(30),
3863 	/* PMCCFILTR */
3864 	{ CP15_PMU_SYS_REG(DIRECT, 0, 14, 15, 7), .access = access_pmu_evtyper },
3865 
3866 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
3867 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
3868 
3869 	/* CCSIDR2 */
3870 	{ Op1(1), CRn( 0), CRm( 0),  Op2(2), undef_access },
3871 
3872 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
3873 };
3874 
3875 static const struct sys_reg_desc cp15_64_regs[] = {
3876 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
3877 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0, 9, 0), .access = access_pmu_evcntr },
3878 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
3879 	{ SYS_DESC(SYS_AARCH32_CNTPCT),	      access_arch_timer },
3880 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
3881 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
3882 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
3883 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
3884 	{ SYS_DESC(SYS_AARCH32_CNTPCTSS),     access_arch_timer },
3885 };
3886 
3887 static bool check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
3888 			       bool is_32)
3889 {
3890 	unsigned int i;
3891 
3892 	for (i = 0; i < n; i++) {
3893 		if (!is_32 && table[i].reg && !table[i].reset) {
3894 			kvm_err("sys_reg table %pS entry %d (%s) lacks reset\n",
3895 				&table[i], i, table[i].name);
3896 			return false;
3897 		}
3898 
3899 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
3900 			kvm_err("sys_reg table %pS entry %d (%s -> %s) out of order\n",
3901 				&table[i], i, table[i - 1].name, table[i].name);
3902 			return false;
3903 		}
3904 	}
3905 
3906 	return true;
3907 }
3908 
3909 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
3910 {
3911 	kvm_inject_undefined(vcpu);
3912 	return 1;
3913 }
3914 
3915 static void perform_access(struct kvm_vcpu *vcpu,
3916 			   struct sys_reg_params *params,
3917 			   const struct sys_reg_desc *r)
3918 {
3919 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
3920 
3921 	/* Check for regs disabled by runtime config */
3922 	if (sysreg_hidden(vcpu, r)) {
3923 		kvm_inject_undefined(vcpu);
3924 		return;
3925 	}
3926 
3927 	/*
3928 	 * Not having an accessor means that we have configured a trap
3929 	 * that we don't know how to handle. This certainly qualifies
3930 	 * as a gross bug that should be fixed right away.
3931 	 */
3932 	BUG_ON(!r->access);
3933 
3934 	/* Skip instruction if instructed so */
3935 	if (likely(r->access(vcpu, params, r)))
3936 		kvm_incr_pc(vcpu);
3937 }
3938 
3939 /*
3940  * emulate_cp --  tries to match a sys_reg access in a handling table, and
3941  *                call the corresponding trap handler.
3942  *
3943  * @params: pointer to the descriptor of the access
3944  * @table: array of trap descriptors
3945  * @num: size of the trap descriptor array
3946  *
3947  * Return true if the access has been handled, false if not.
3948  */
3949 static bool emulate_cp(struct kvm_vcpu *vcpu,
3950 		       struct sys_reg_params *params,
3951 		       const struct sys_reg_desc *table,
3952 		       size_t num)
3953 {
3954 	const struct sys_reg_desc *r;
3955 
3956 	if (!table)
3957 		return false;	/* Not handled */
3958 
3959 	r = find_reg(params, table, num);
3960 
3961 	if (r) {
3962 		perform_access(vcpu, params, r);
3963 		return true;
3964 	}
3965 
3966 	/* Not handled */
3967 	return false;
3968 }
3969 
3970 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
3971 				struct sys_reg_params *params)
3972 {
3973 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
3974 	int cp = -1;
3975 
3976 	switch (esr_ec) {
3977 	case ESR_ELx_EC_CP15_32:
3978 	case ESR_ELx_EC_CP15_64:
3979 		cp = 15;
3980 		break;
3981 	case ESR_ELx_EC_CP14_MR:
3982 	case ESR_ELx_EC_CP14_64:
3983 		cp = 14;
3984 		break;
3985 	default:
3986 		WARN_ON(1);
3987 	}
3988 
3989 	print_sys_reg_msg(params,
3990 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
3991 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
3992 	kvm_inject_undefined(vcpu);
3993 }
3994 
3995 /**
3996  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
3997  * @vcpu: The VCPU pointer
3998  * @global: &struct sys_reg_desc
3999  * @nr_global: size of the @global array
4000  */
4001 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
4002 			    const struct sys_reg_desc *global,
4003 			    size_t nr_global)
4004 {
4005 	struct sys_reg_params params;
4006 	u64 esr = kvm_vcpu_get_esr(vcpu);
4007 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4008 	int Rt2 = (esr >> 10) & 0x1f;
4009 
4010 	params.CRm = (esr >> 1) & 0xf;
4011 	params.is_write = ((esr & 1) == 0);
4012 
4013 	params.Op0 = 0;
4014 	params.Op1 = (esr >> 16) & 0xf;
4015 	params.Op2 = 0;
4016 	params.CRn = 0;
4017 
4018 	/*
4019 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
4020 	 * backends between AArch32 and AArch64, we get away with it.
4021 	 */
4022 	if (params.is_write) {
4023 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
4024 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
4025 	}
4026 
4027 	/*
4028 	 * If the table contains a handler, handle the
4029 	 * potential register operation in the case of a read and return
4030 	 * with success.
4031 	 */
4032 	if (emulate_cp(vcpu, &params, global, nr_global)) {
4033 		/* Split up the value between registers for the read side */
4034 		if (!params.is_write) {
4035 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
4036 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
4037 		}
4038 
4039 		return 1;
4040 	}
4041 
4042 	unhandled_cp_access(vcpu, &params);
4043 	return 1;
4044 }
4045 
4046 static bool emulate_sys_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *params);
4047 
4048 /*
4049  * The CP10 ID registers are architecturally mapped to AArch64 feature
4050  * registers. Abuse that fact so we can rely on the AArch64 handler for accesses
4051  * from AArch32.
4052  */
4053 static bool kvm_esr_cp10_id_to_sys64(u64 esr, struct sys_reg_params *params)
4054 {
4055 	u8 reg_id = (esr >> 10) & 0xf;
4056 	bool valid;
4057 
4058 	params->is_write = ((esr & 1) == 0);
4059 	params->Op0 = 3;
4060 	params->Op1 = 0;
4061 	params->CRn = 0;
4062 	params->CRm = 3;
4063 
4064 	/* CP10 ID registers are read-only */
4065 	valid = !params->is_write;
4066 
4067 	switch (reg_id) {
4068 	/* MVFR0 */
4069 	case 0b0111:
4070 		params->Op2 = 0;
4071 		break;
4072 	/* MVFR1 */
4073 	case 0b0110:
4074 		params->Op2 = 1;
4075 		break;
4076 	/* MVFR2 */
4077 	case 0b0101:
4078 		params->Op2 = 2;
4079 		break;
4080 	default:
4081 		valid = false;
4082 	}
4083 
4084 	if (valid)
4085 		return true;
4086 
4087 	kvm_pr_unimpl("Unhandled cp10 register %s: %u\n",
4088 		      params->is_write ? "write" : "read", reg_id);
4089 	return false;
4090 }
4091 
4092 /**
4093  * kvm_handle_cp10_id() - Handles a VMRS trap on guest access to a 'Media and
4094  *			  VFP Register' from AArch32.
4095  * @vcpu: The vCPU pointer
4096  *
4097  * MVFR{0-2} are architecturally mapped to the AArch64 MVFR{0-2}_EL1 registers.
4098  * Work out the correct AArch64 system register encoding and reroute to the
4099  * AArch64 system register emulation.
4100  */
4101 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu)
4102 {
4103 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4104 	u64 esr = kvm_vcpu_get_esr(vcpu);
4105 	struct sys_reg_params params;
4106 
4107 	/* UNDEF on any unhandled register access */
4108 	if (!kvm_esr_cp10_id_to_sys64(esr, &params)) {
4109 		kvm_inject_undefined(vcpu);
4110 		return 1;
4111 	}
4112 
4113 	if (emulate_sys_reg(vcpu, &params))
4114 		vcpu_set_reg(vcpu, Rt, params.regval);
4115 
4116 	return 1;
4117 }
4118 
4119 /**
4120  * kvm_emulate_cp15_id_reg() - Handles an MRC trap on a guest CP15 access where
4121  *			       CRn=0, which corresponds to the AArch32 feature
4122  *			       registers.
4123  * @vcpu: the vCPU pointer
4124  * @params: the system register access parameters.
4125  *
4126  * Our cp15 system register tables do not enumerate the AArch32 feature
4127  * registers. Conveniently, our AArch64 table does, and the AArch32 system
4128  * register encoding can be trivially remapped into the AArch64 for the feature
4129  * registers: Append op0=3, leaving op1, CRn, CRm, and op2 the same.
4130  *
4131  * According to DDI0487G.b G7.3.1, paragraph "Behavior of VMSAv8-32 32-bit
4132  * System registers with (coproc=0b1111, CRn==c0)", read accesses from this
4133  * range are either UNKNOWN or RES0. Rerouting remains architectural as we
4134  * treat undefined registers in this range as RAZ.
4135  */
4136 static int kvm_emulate_cp15_id_reg(struct kvm_vcpu *vcpu,
4137 				   struct sys_reg_params *params)
4138 {
4139 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4140 
4141 	/* Treat impossible writes to RO registers as UNDEFINED */
4142 	if (params->is_write) {
4143 		unhandled_cp_access(vcpu, params);
4144 		return 1;
4145 	}
4146 
4147 	params->Op0 = 3;
4148 
4149 	/*
4150 	 * All registers where CRm > 3 are known to be UNKNOWN/RAZ from AArch32.
4151 	 * Avoid conflicting with future expansion of AArch64 feature registers
4152 	 * and simply treat them as RAZ here.
4153 	 */
4154 	if (params->CRm > 3)
4155 		params->regval = 0;
4156 	else if (!emulate_sys_reg(vcpu, params))
4157 		return 1;
4158 
4159 	vcpu_set_reg(vcpu, Rt, params->regval);
4160 	return 1;
4161 }
4162 
4163 /**
4164  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
4165  * @vcpu: The VCPU pointer
4166  * @params: &struct sys_reg_params
4167  * @global: &struct sys_reg_desc
4168  * @nr_global: size of the @global array
4169  */
4170 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
4171 			    struct sys_reg_params *params,
4172 			    const struct sys_reg_desc *global,
4173 			    size_t nr_global)
4174 {
4175 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
4176 
4177 	params->regval = vcpu_get_reg(vcpu, Rt);
4178 
4179 	if (emulate_cp(vcpu, params, global, nr_global)) {
4180 		if (!params->is_write)
4181 			vcpu_set_reg(vcpu, Rt, params->regval);
4182 		return 1;
4183 	}
4184 
4185 	unhandled_cp_access(vcpu, params);
4186 	return 1;
4187 }
4188 
4189 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
4190 {
4191 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
4192 }
4193 
4194 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
4195 {
4196 	struct sys_reg_params params;
4197 
4198 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
4199 
4200 	/*
4201 	 * Certain AArch32 ID registers are handled by rerouting to the AArch64
4202 	 * system register table. Registers in the ID range where CRm=0 are
4203 	 * excluded from this scheme as they do not trivially map into AArch64
4204 	 * system register encodings.
4205 	 */
4206 	if (params.Op1 == 0 && params.CRn == 0 && params.CRm)
4207 		return kvm_emulate_cp15_id_reg(vcpu, &params);
4208 
4209 	return kvm_handle_cp_32(vcpu, &params, cp15_regs, ARRAY_SIZE(cp15_regs));
4210 }
4211 
4212 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
4213 {
4214 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
4215 }
4216 
4217 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
4218 {
4219 	struct sys_reg_params params;
4220 
4221 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
4222 
4223 	return kvm_handle_cp_32(vcpu, &params, cp14_regs, ARRAY_SIZE(cp14_regs));
4224 }
4225 
4226 /**
4227  * emulate_sys_reg - Emulate a guest access to an AArch64 system register
4228  * @vcpu: The VCPU pointer
4229  * @params: Decoded system register parameters
4230  *
4231  * Return: true if the system register access was successful, false otherwise.
4232  */
4233 static bool emulate_sys_reg(struct kvm_vcpu *vcpu,
4234 			    struct sys_reg_params *params)
4235 {
4236 	const struct sys_reg_desc *r;
4237 
4238 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4239 	if (likely(r)) {
4240 		perform_access(vcpu, params, r);
4241 		return true;
4242 	}
4243 
4244 	print_sys_reg_msg(params,
4245 			  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
4246 			  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
4247 	kvm_inject_undefined(vcpu);
4248 
4249 	return false;
4250 }
4251 
4252 static const struct sys_reg_desc *idregs_debug_find(struct kvm *kvm, u8 pos)
4253 {
4254 	unsigned long i, idreg_idx = 0;
4255 
4256 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4257 		const struct sys_reg_desc *r = &sys_reg_descs[i];
4258 
4259 		if (!is_vm_ftr_id_reg(reg_to_encoding(r)))
4260 			continue;
4261 
4262 		if (idreg_idx == pos)
4263 			return r;
4264 
4265 		idreg_idx++;
4266 	}
4267 
4268 	return NULL;
4269 }
4270 
4271 static void *idregs_debug_start(struct seq_file *s, loff_t *pos)
4272 {
4273 	struct kvm *kvm = s->private;
4274 	u8 *iter;
4275 
4276 	mutex_lock(&kvm->arch.config_lock);
4277 
4278 	iter = &kvm->arch.idreg_debugfs_iter;
4279 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags) &&
4280 	    *iter == (u8)~0) {
4281 		*iter = *pos;
4282 		if (!idregs_debug_find(kvm, *iter))
4283 			iter = NULL;
4284 	} else {
4285 		iter = ERR_PTR(-EBUSY);
4286 	}
4287 
4288 	mutex_unlock(&kvm->arch.config_lock);
4289 
4290 	return iter;
4291 }
4292 
4293 static void *idregs_debug_next(struct seq_file *s, void *v, loff_t *pos)
4294 {
4295 	struct kvm *kvm = s->private;
4296 
4297 	(*pos)++;
4298 
4299 	if (idregs_debug_find(kvm, kvm->arch.idreg_debugfs_iter + 1)) {
4300 		kvm->arch.idreg_debugfs_iter++;
4301 
4302 		return &kvm->arch.idreg_debugfs_iter;
4303 	}
4304 
4305 	return NULL;
4306 }
4307 
4308 static void idregs_debug_stop(struct seq_file *s, void *v)
4309 {
4310 	struct kvm *kvm = s->private;
4311 
4312 	if (IS_ERR(v))
4313 		return;
4314 
4315 	mutex_lock(&kvm->arch.config_lock);
4316 
4317 	kvm->arch.idreg_debugfs_iter = ~0;
4318 
4319 	mutex_unlock(&kvm->arch.config_lock);
4320 }
4321 
4322 static int idregs_debug_show(struct seq_file *s, void *v)
4323 {
4324 	const struct sys_reg_desc *desc;
4325 	struct kvm *kvm = s->private;
4326 
4327 	desc = idregs_debug_find(kvm, kvm->arch.idreg_debugfs_iter);
4328 
4329 	if (!desc->name)
4330 		return 0;
4331 
4332 	seq_printf(s, "%20s:\t%016llx\n",
4333 		   desc->name, kvm_read_vm_id_reg(kvm, reg_to_encoding(desc)));
4334 
4335 	return 0;
4336 }
4337 
4338 static const struct seq_operations idregs_debug_sops = {
4339 	.start	= idregs_debug_start,
4340 	.next	= idregs_debug_next,
4341 	.stop	= idregs_debug_stop,
4342 	.show	= idregs_debug_show,
4343 };
4344 
4345 DEFINE_SEQ_ATTRIBUTE(idregs_debug);
4346 
4347 void kvm_sys_regs_create_debugfs(struct kvm *kvm)
4348 {
4349 	kvm->arch.idreg_debugfs_iter = ~0;
4350 
4351 	debugfs_create_file("idregs", 0444, kvm->debugfs_dentry, kvm,
4352 			    &idregs_debug_fops);
4353 }
4354 
4355 static void reset_vm_ftr_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *reg)
4356 {
4357 	u32 id = reg_to_encoding(reg);
4358 	struct kvm *kvm = vcpu->kvm;
4359 
4360 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags))
4361 		return;
4362 
4363 	kvm_set_vm_id_reg(kvm, id, reg->reset(vcpu, reg));
4364 }
4365 
4366 static void reset_vcpu_ftr_id_reg(struct kvm_vcpu *vcpu,
4367 				  const struct sys_reg_desc *reg)
4368 {
4369 	if (kvm_vcpu_initialized(vcpu))
4370 		return;
4371 
4372 	reg->reset(vcpu, reg);
4373 }
4374 
4375 /**
4376  * kvm_reset_sys_regs - sets system registers to reset value
4377  * @vcpu: The VCPU pointer
4378  *
4379  * This function finds the right table above and sets the registers on the
4380  * virtual CPU struct to their architecturally defined reset values.
4381  */
4382 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
4383 {
4384 	struct kvm *kvm = vcpu->kvm;
4385 	unsigned long i;
4386 
4387 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4388 		const struct sys_reg_desc *r = &sys_reg_descs[i];
4389 
4390 		if (!r->reset)
4391 			continue;
4392 
4393 		if (is_vm_ftr_id_reg(reg_to_encoding(r)))
4394 			reset_vm_ftr_id_reg(vcpu, r);
4395 		else if (is_vcpu_ftr_id_reg(reg_to_encoding(r)))
4396 			reset_vcpu_ftr_id_reg(vcpu, r);
4397 		else
4398 			r->reset(vcpu, r);
4399 	}
4400 
4401 	set_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags);
4402 }
4403 
4404 /**
4405  * kvm_handle_sys_reg -- handles a system instruction or mrs/msr instruction
4406  *			 trap on a guest execution
4407  * @vcpu: The VCPU pointer
4408  */
4409 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
4410 {
4411 	const struct sys_reg_desc *desc = NULL;
4412 	struct sys_reg_params params;
4413 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
4414 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4415 	int sr_idx;
4416 
4417 	trace_kvm_handle_sys_reg(esr);
4418 
4419 	if (triage_sysreg_trap(vcpu, &sr_idx))
4420 		return 1;
4421 
4422 	params = esr_sys64_to_params(esr);
4423 	params.regval = vcpu_get_reg(vcpu, Rt);
4424 
4425 	/* System registers have Op0=={2,3}, as per DDI487 J.a C5.1.2 */
4426 	if (params.Op0 == 2 || params.Op0 == 3)
4427 		desc = &sys_reg_descs[sr_idx];
4428 	else
4429 		desc = &sys_insn_descs[sr_idx];
4430 
4431 	perform_access(vcpu, &params, desc);
4432 
4433 	/* Read from system register? */
4434 	if (!params.is_write &&
4435 	    (params.Op0 == 2 || params.Op0 == 3))
4436 		vcpu_set_reg(vcpu, Rt, params.regval);
4437 
4438 	return 1;
4439 }
4440 
4441 /******************************************************************************
4442  * Userspace API
4443  *****************************************************************************/
4444 
4445 static bool index_to_params(u64 id, struct sys_reg_params *params)
4446 {
4447 	switch (id & KVM_REG_SIZE_MASK) {
4448 	case KVM_REG_SIZE_U64:
4449 		/* Any unused index bits means it's not valid. */
4450 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
4451 			      | KVM_REG_ARM_COPROC_MASK
4452 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
4453 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
4454 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
4455 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
4456 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
4457 			return false;
4458 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
4459 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
4460 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
4461 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
4462 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
4463 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
4464 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
4465 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
4466 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
4467 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
4468 		return true;
4469 	default:
4470 		return false;
4471 	}
4472 }
4473 
4474 const struct sys_reg_desc *get_reg_by_id(u64 id,
4475 					 const struct sys_reg_desc table[],
4476 					 unsigned int num)
4477 {
4478 	struct sys_reg_params params;
4479 
4480 	if (!index_to_params(id, &params))
4481 		return NULL;
4482 
4483 	return find_reg(&params, table, num);
4484 }
4485 
4486 /* Decode an index value, and find the sys_reg_desc entry. */
4487 static const struct sys_reg_desc *
4488 id_to_sys_reg_desc(struct kvm_vcpu *vcpu, u64 id,
4489 		   const struct sys_reg_desc table[], unsigned int num)
4490 
4491 {
4492 	const struct sys_reg_desc *r;
4493 
4494 	/* We only do sys_reg for now. */
4495 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
4496 		return NULL;
4497 
4498 	r = get_reg_by_id(id, table, num);
4499 
4500 	/* Not saved in the sys_reg array and not otherwise accessible? */
4501 	if (r && (!(r->reg || r->get_user) || sysreg_hidden(vcpu, r)))
4502 		r = NULL;
4503 
4504 	return r;
4505 }
4506 
4507 /*
4508  * These are the invariant sys_reg registers: we let the guest see the
4509  * host versions of these, so they're part of the guest state.
4510  *
4511  * A future CPU may provide a mechanism to present different values to
4512  * the guest, or a future kvm may trap them.
4513  */
4514 
4515 #define FUNCTION_INVARIANT(reg)						\
4516 	static u64 reset_##reg(struct kvm_vcpu *v,			\
4517 			       const struct sys_reg_desc *r)		\
4518 	{								\
4519 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
4520 		return ((struct sys_reg_desc *)r)->val;			\
4521 	}
4522 
4523 FUNCTION_INVARIANT(midr_el1)
4524 FUNCTION_INVARIANT(revidr_el1)
4525 FUNCTION_INVARIANT(aidr_el1)
4526 
4527 /* ->val is filled in by kvm_sys_reg_table_init() */
4528 static struct sys_reg_desc invariant_sys_regs[] __ro_after_init = {
4529 	{ SYS_DESC(SYS_MIDR_EL1), NULL, reset_midr_el1 },
4530 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, reset_revidr_el1 },
4531 	{ SYS_DESC(SYS_AIDR_EL1), NULL, reset_aidr_el1 },
4532 };
4533 
4534 static int get_invariant_sys_reg(u64 id, u64 __user *uaddr)
4535 {
4536 	const struct sys_reg_desc *r;
4537 
4538 	r = get_reg_by_id(id, invariant_sys_regs,
4539 			  ARRAY_SIZE(invariant_sys_regs));
4540 	if (!r)
4541 		return -ENOENT;
4542 
4543 	return put_user(r->val, uaddr);
4544 }
4545 
4546 static int set_invariant_sys_reg(u64 id, u64 __user *uaddr)
4547 {
4548 	const struct sys_reg_desc *r;
4549 	u64 val;
4550 
4551 	r = get_reg_by_id(id, invariant_sys_regs,
4552 			  ARRAY_SIZE(invariant_sys_regs));
4553 	if (!r)
4554 		return -ENOENT;
4555 
4556 	if (get_user(val, uaddr))
4557 		return -EFAULT;
4558 
4559 	/* This is what we mean by invariant: you can't change it. */
4560 	if (r->val != val)
4561 		return -EINVAL;
4562 
4563 	return 0;
4564 }
4565 
4566 static int demux_c15_get(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
4567 {
4568 	u32 val;
4569 	u32 __user *uval = uaddr;
4570 
4571 	/* Fail if we have unknown bits set. */
4572 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
4573 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
4574 		return -ENOENT;
4575 
4576 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
4577 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
4578 		if (KVM_REG_SIZE(id) != 4)
4579 			return -ENOENT;
4580 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
4581 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
4582 		if (val >= CSSELR_MAX)
4583 			return -ENOENT;
4584 
4585 		return put_user(get_ccsidr(vcpu, val), uval);
4586 	default:
4587 		return -ENOENT;
4588 	}
4589 }
4590 
4591 static int demux_c15_set(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
4592 {
4593 	u32 val, newval;
4594 	u32 __user *uval = uaddr;
4595 
4596 	/* Fail if we have unknown bits set. */
4597 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
4598 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
4599 		return -ENOENT;
4600 
4601 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
4602 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
4603 		if (KVM_REG_SIZE(id) != 4)
4604 			return -ENOENT;
4605 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
4606 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
4607 		if (val >= CSSELR_MAX)
4608 			return -ENOENT;
4609 
4610 		if (get_user(newval, uval))
4611 			return -EFAULT;
4612 
4613 		return set_ccsidr(vcpu, val, newval);
4614 	default:
4615 		return -ENOENT;
4616 	}
4617 }
4618 
4619 int kvm_sys_reg_get_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
4620 			 const struct sys_reg_desc table[], unsigned int num)
4621 {
4622 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
4623 	const struct sys_reg_desc *r;
4624 	u64 val;
4625 	int ret;
4626 
4627 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
4628 	if (!r || sysreg_hidden(vcpu, r))
4629 		return -ENOENT;
4630 
4631 	if (r->get_user) {
4632 		ret = (r->get_user)(vcpu, r, &val);
4633 	} else {
4634 		val = __vcpu_sys_reg(vcpu, r->reg);
4635 		ret = 0;
4636 	}
4637 
4638 	if (!ret)
4639 		ret = put_user(val, uaddr);
4640 
4641 	return ret;
4642 }
4643 
4644 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
4645 {
4646 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
4647 	int err;
4648 
4649 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
4650 		return demux_c15_get(vcpu, reg->id, uaddr);
4651 
4652 	err = get_invariant_sys_reg(reg->id, uaddr);
4653 	if (err != -ENOENT)
4654 		return err;
4655 
4656 	return kvm_sys_reg_get_user(vcpu, reg,
4657 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4658 }
4659 
4660 int kvm_sys_reg_set_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
4661 			 const struct sys_reg_desc table[], unsigned int num)
4662 {
4663 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
4664 	const struct sys_reg_desc *r;
4665 	u64 val;
4666 	int ret;
4667 
4668 	if (get_user(val, uaddr))
4669 		return -EFAULT;
4670 
4671 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
4672 	if (!r || sysreg_hidden(vcpu, r))
4673 		return -ENOENT;
4674 
4675 	if (sysreg_user_write_ignore(vcpu, r))
4676 		return 0;
4677 
4678 	if (r->set_user) {
4679 		ret = (r->set_user)(vcpu, r, val);
4680 	} else {
4681 		__vcpu_sys_reg(vcpu, r->reg) = val;
4682 		ret = 0;
4683 	}
4684 
4685 	return ret;
4686 }
4687 
4688 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
4689 {
4690 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
4691 	int err;
4692 
4693 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
4694 		return demux_c15_set(vcpu, reg->id, uaddr);
4695 
4696 	err = set_invariant_sys_reg(reg->id, uaddr);
4697 	if (err != -ENOENT)
4698 		return err;
4699 
4700 	return kvm_sys_reg_set_user(vcpu, reg,
4701 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4702 }
4703 
4704 static unsigned int num_demux_regs(void)
4705 {
4706 	return CSSELR_MAX;
4707 }
4708 
4709 static int write_demux_regids(u64 __user *uindices)
4710 {
4711 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
4712 	unsigned int i;
4713 
4714 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
4715 	for (i = 0; i < CSSELR_MAX; i++) {
4716 		if (put_user(val | i, uindices))
4717 			return -EFAULT;
4718 		uindices++;
4719 	}
4720 	return 0;
4721 }
4722 
4723 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
4724 {
4725 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
4726 		KVM_REG_ARM64_SYSREG |
4727 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
4728 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
4729 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
4730 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
4731 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
4732 }
4733 
4734 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
4735 {
4736 	if (!*uind)
4737 		return true;
4738 
4739 	if (put_user(sys_reg_to_index(reg), *uind))
4740 		return false;
4741 
4742 	(*uind)++;
4743 	return true;
4744 }
4745 
4746 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
4747 			    const struct sys_reg_desc *rd,
4748 			    u64 __user **uind,
4749 			    unsigned int *total)
4750 {
4751 	/*
4752 	 * Ignore registers we trap but don't save,
4753 	 * and for which no custom user accessor is provided.
4754 	 */
4755 	if (!(rd->reg || rd->get_user))
4756 		return 0;
4757 
4758 	if (sysreg_hidden(vcpu, rd))
4759 		return 0;
4760 
4761 	if (!copy_reg_to_user(rd, uind))
4762 		return -EFAULT;
4763 
4764 	(*total)++;
4765 	return 0;
4766 }
4767 
4768 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
4769 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
4770 {
4771 	const struct sys_reg_desc *i2, *end2;
4772 	unsigned int total = 0;
4773 	int err;
4774 
4775 	i2 = sys_reg_descs;
4776 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
4777 
4778 	while (i2 != end2) {
4779 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
4780 		if (err)
4781 			return err;
4782 	}
4783 	return total;
4784 }
4785 
4786 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
4787 {
4788 	return ARRAY_SIZE(invariant_sys_regs)
4789 		+ num_demux_regs()
4790 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
4791 }
4792 
4793 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
4794 {
4795 	unsigned int i;
4796 	int err;
4797 
4798 	/* Then give them all the invariant registers' indices. */
4799 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
4800 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
4801 			return -EFAULT;
4802 		uindices++;
4803 	}
4804 
4805 	err = walk_sys_regs(vcpu, uindices);
4806 	if (err < 0)
4807 		return err;
4808 	uindices += err;
4809 
4810 	return write_demux_regids(uindices);
4811 }
4812 
4813 #define KVM_ARM_FEATURE_ID_RANGE_INDEX(r)			\
4814 	KVM_ARM_FEATURE_ID_RANGE_IDX(sys_reg_Op0(r),		\
4815 		sys_reg_Op1(r),					\
4816 		sys_reg_CRn(r),					\
4817 		sys_reg_CRm(r),					\
4818 		sys_reg_Op2(r))
4819 
4820 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm, struct reg_mask_range *range)
4821 {
4822 	const void *zero_page = page_to_virt(ZERO_PAGE(0));
4823 	u64 __user *masks = (u64 __user *)range->addr;
4824 
4825 	/* Only feature id range is supported, reserved[13] must be zero. */
4826 	if (range->range ||
4827 	    memcmp(range->reserved, zero_page, sizeof(range->reserved)))
4828 		return -EINVAL;
4829 
4830 	/* Wipe the whole thing first */
4831 	if (clear_user(masks, KVM_ARM_FEATURE_ID_RANGE_SIZE * sizeof(__u64)))
4832 		return -EFAULT;
4833 
4834 	for (int i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4835 		const struct sys_reg_desc *reg = &sys_reg_descs[i];
4836 		u32 encoding = reg_to_encoding(reg);
4837 		u64 val;
4838 
4839 		if (!is_feature_id_reg(encoding) || !reg->set_user)
4840 			continue;
4841 
4842 		if (!reg->val ||
4843 		    (is_aa32_id_reg(encoding) && !kvm_supports_32bit_el0())) {
4844 			continue;
4845 		}
4846 		val = reg->val;
4847 
4848 		if (put_user(val, (masks + KVM_ARM_FEATURE_ID_RANGE_INDEX(encoding))))
4849 			return -EFAULT;
4850 	}
4851 
4852 	return 0;
4853 }
4854 
4855 static void vcpu_set_hcr(struct kvm_vcpu *vcpu)
4856 {
4857 	struct kvm *kvm = vcpu->kvm;
4858 
4859 	if (has_vhe() || has_hvhe())
4860 		vcpu->arch.hcr_el2 |= HCR_E2H;
4861 	if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN)) {
4862 		/* route synchronous external abort exceptions to EL2 */
4863 		vcpu->arch.hcr_el2 |= HCR_TEA;
4864 		/* trap error record accesses */
4865 		vcpu->arch.hcr_el2 |= HCR_TERR;
4866 	}
4867 
4868 	if (cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
4869 		vcpu->arch.hcr_el2 |= HCR_FWB;
4870 
4871 	if (cpus_have_final_cap(ARM64_HAS_EVT) &&
4872 	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE) &&
4873 	    kvm_read_vm_id_reg(kvm, SYS_CTR_EL0) == read_sanitised_ftr_reg(SYS_CTR_EL0))
4874 		vcpu->arch.hcr_el2 |= HCR_TID4;
4875 	else
4876 		vcpu->arch.hcr_el2 |= HCR_TID2;
4877 
4878 	if (vcpu_el1_is_32bit(vcpu))
4879 		vcpu->arch.hcr_el2 &= ~HCR_RW;
4880 
4881 	if (kvm_has_mte(vcpu->kvm))
4882 		vcpu->arch.hcr_el2 |= HCR_ATA;
4883 
4884 	/*
4885 	 * In the absence of FGT, we cannot independently trap TLBI
4886 	 * Range instructions. This isn't great, but trapping all
4887 	 * TLBIs would be far worse. Live with it...
4888 	 */
4889 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
4890 		vcpu->arch.hcr_el2 |= HCR_TTLBOS;
4891 }
4892 
4893 void kvm_calculate_traps(struct kvm_vcpu *vcpu)
4894 {
4895 	struct kvm *kvm = vcpu->kvm;
4896 
4897 	mutex_lock(&kvm->arch.config_lock);
4898 	vcpu_set_hcr(vcpu);
4899 	vcpu_set_ich_hcr(vcpu);
4900 
4901 	if (cpus_have_final_cap(ARM64_HAS_HCX)) {
4902 		/*
4903 		 * In general, all HCRX_EL2 bits are gated by a feature.
4904 		 * The only reason we can set SMPME without checking any
4905 		 * feature is that its effects are not directly observable
4906 		 * from the guest.
4907 		 */
4908 		vcpu->arch.hcrx_el2 = HCRX_EL2_SMPME;
4909 
4910 		if (kvm_has_feat(kvm, ID_AA64ISAR2_EL1, MOPS, IMP))
4911 			vcpu->arch.hcrx_el2 |= (HCRX_EL2_MSCEn | HCRX_EL2_MCE2);
4912 
4913 		if (kvm_has_tcr2(kvm))
4914 			vcpu->arch.hcrx_el2 |= HCRX_EL2_TCR2En;
4915 
4916 		if (kvm_has_fpmr(kvm))
4917 			vcpu->arch.hcrx_el2 |= HCRX_EL2_EnFPM;
4918 	}
4919 
4920 	if (test_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags))
4921 		goto out;
4922 
4923 	kvm->arch.fgu[HFGxTR_GROUP] = (HFGxTR_EL2_nAMAIR2_EL1		|
4924 				       HFGxTR_EL2_nMAIR2_EL1		|
4925 				       HFGxTR_EL2_nS2POR_EL1		|
4926 				       HFGxTR_EL2_nACCDATA_EL1		|
4927 				       HFGxTR_EL2_nSMPRI_EL1_MASK	|
4928 				       HFGxTR_EL2_nTPIDR2_EL0_MASK);
4929 
4930 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
4931 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_TLBIRVAALE1OS|
4932 						HFGITR_EL2_TLBIRVALE1OS	|
4933 						HFGITR_EL2_TLBIRVAAE1OS	|
4934 						HFGITR_EL2_TLBIRVAE1OS	|
4935 						HFGITR_EL2_TLBIVAALE1OS	|
4936 						HFGITR_EL2_TLBIVALE1OS	|
4937 						HFGITR_EL2_TLBIVAAE1OS	|
4938 						HFGITR_EL2_TLBIASIDE1OS	|
4939 						HFGITR_EL2_TLBIVAE1OS	|
4940 						HFGITR_EL2_TLBIVMALLE1OS);
4941 
4942 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
4943 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_TLBIRVAALE1	|
4944 						HFGITR_EL2_TLBIRVALE1	|
4945 						HFGITR_EL2_TLBIRVAAE1	|
4946 						HFGITR_EL2_TLBIRVAE1	|
4947 						HFGITR_EL2_TLBIRVAALE1IS|
4948 						HFGITR_EL2_TLBIRVALE1IS	|
4949 						HFGITR_EL2_TLBIRVAAE1IS	|
4950 						HFGITR_EL2_TLBIRVAE1IS	|
4951 						HFGITR_EL2_TLBIRVAALE1OS|
4952 						HFGITR_EL2_TLBIRVALE1OS	|
4953 						HFGITR_EL2_TLBIRVAAE1OS	|
4954 						HFGITR_EL2_TLBIRVAE1OS);
4955 
4956 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, ATS1A, IMP))
4957 		kvm->arch.fgu[HFGITR_GROUP] |= HFGITR_EL2_ATS1E1A;
4958 
4959 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, PAN, PAN2))
4960 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_ATS1E1RP |
4961 						HFGITR_EL2_ATS1E1WP);
4962 
4963 	if (!kvm_has_s1pie(kvm))
4964 		kvm->arch.fgu[HFGxTR_GROUP] |= (HFGxTR_EL2_nPIRE0_EL1 |
4965 						HFGxTR_EL2_nPIR_EL1);
4966 
4967 	if (!kvm_has_s1poe(kvm))
4968 		kvm->arch.fgu[HFGxTR_GROUP] |= (HFGxTR_EL2_nPOR_EL1 |
4969 						HFGxTR_EL2_nPOR_EL0);
4970 
4971 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, AMU, IMP))
4972 		kvm->arch.fgu[HAFGRTR_GROUP] |= ~(HAFGRTR_EL2_RES0 |
4973 						  HAFGRTR_EL2_RES1);
4974 
4975 	set_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags);
4976 out:
4977 	mutex_unlock(&kvm->arch.config_lock);
4978 }
4979 
4980 /*
4981  * Perform last adjustments to the ID registers that are implied by the
4982  * configuration outside of the ID regs themselves, as well as any
4983  * initialisation that directly depend on these ID registers (such as
4984  * RES0/RES1 behaviours). This is not the place to configure traps though.
4985  *
4986  * Because this can be called once per CPU, changes must be idempotent.
4987  */
4988 int kvm_finalize_sys_regs(struct kvm_vcpu *vcpu)
4989 {
4990 	struct kvm *kvm = vcpu->kvm;
4991 
4992 	guard(mutex)(&kvm->arch.config_lock);
4993 
4994 	if (!(static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif) &&
4995 	      irqchip_in_kernel(kvm) &&
4996 	      kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V3)) {
4997 		kvm->arch.id_regs[IDREG_IDX(SYS_ID_AA64PFR0_EL1)] &= ~ID_AA64PFR0_EL1_GIC_MASK;
4998 		kvm->arch.id_regs[IDREG_IDX(SYS_ID_PFR1_EL1)] &= ~ID_PFR1_EL1_GIC_MASK;
4999 	}
5000 
5001 	if (vcpu_has_nv(vcpu)) {
5002 		int ret = kvm_init_nv_sysregs(kvm);
5003 		if (ret)
5004 			return ret;
5005 	}
5006 
5007 	return 0;
5008 }
5009 
5010 int __init kvm_sys_reg_table_init(void)
5011 {
5012 	bool valid = true;
5013 	unsigned int i;
5014 	int ret = 0;
5015 
5016 	/* Make sure tables are unique and in order. */
5017 	valid &= check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false);
5018 	valid &= check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true);
5019 	valid &= check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true);
5020 	valid &= check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true);
5021 	valid &= check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true);
5022 	valid &= check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false);
5023 	valid &= check_sysreg_table(sys_insn_descs, ARRAY_SIZE(sys_insn_descs), false);
5024 
5025 	if (!valid)
5026 		return -EINVAL;
5027 
5028 	/* We abuse the reset function to overwrite the table itself. */
5029 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
5030 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
5031 
5032 	ret = populate_nv_trap_config();
5033 
5034 	for (i = 0; !ret && i < ARRAY_SIZE(sys_reg_descs); i++)
5035 		ret = populate_sysreg_config(sys_reg_descs + i, i);
5036 
5037 	for (i = 0; !ret && i < ARRAY_SIZE(sys_insn_descs); i++)
5038 		ret = populate_sysreg_config(sys_insn_descs + i, i);
5039 
5040 	return ret;
5041 }
5042